TW200535609A - Disk controller methods and apparatus with improved striping, redundancy operations and interfaces - Google Patents

Abstract

A RAID disk drive controller (FIG. 33) implements disk storage operations, including striping and redundancy operations with multiple disk drives connected via respective SATA ports (520). Configurable data path switch logic (460) provides dynamic configuration of two or more attached drives into one or more arrays. Data transfers are synchronized locally by leveraging the SATA port transport layer FIFO (530). Synchronous transfers allow on-the-fly redundancy (XOR) operations (FIG. 36) for improved performance and reduced hardware complexity. XOR accumulator hardware (FIG. 42-FIG. 43) reduces buffer requirements for multiple DMA channels otherwise required for synchronization, and various narrow and wide striping modes are supported.

Description

200535609 IX. Description of Invention: [Priority Claim] This application claims the priority of US Provisional Patent No. 6G / 553,594 filed on March 12, 2004. The provisional patent case is incorporated herein by reference in its entirety. [Technical field to which the invention belongs] The present invention relates to the field of digital data storage systems, and more specifically relates to the improvement of RAID controllers. [Previous Technology] The hard disk drive is still the main mass storage feature for small to medium-sized computers, which is mainly composed of search time and rotation delay. Limited by 婵 n—100 mechanical parts, the data transmission rate is the data transmission wheel; binocular, ΛΓ_) electronic electricity ... Teng μ per second Γ product and within the range of 50 to 100. One hundred and forty-four groups, per second) One type of system may use multiple magnetic machines, if the required capacity and performance. From a single size ... Gu Xia's most common motivation. -Given a 10 ′ drive that can store data such as Θ 疋; tl_y ... 卞 > n dry-twice as much as a drive 'cause +', compare Figure 1A and Figure 1B. 4 of the given pattern of _ 纟 ## # 度 纟 Yueqiang is less obvious. Each drive will have twice the efficiency of a single drive. On the other hand, 'a system can be configured but it has the data of the first drive-it is true that each drive is the total Λ ί (^ M r * only copying (this is often called: mirroring a lng) "). If any 1 disc player: Er Yu still has-one copy without loss; see section 1 (: Figure (/ in, material A, is a copy of poor material A). ', Bei Yi said, — Make * go ☆ The failed disk drive is replaced immediately and re-established. „The chance of material loss is a failed disk drive and re-established f ', the probability of failure of the replacement machine. This is much lower than any — The magnetic disk rate. The third reason for the failure of the early disk drive is performance. It is: — the use of the system disk array. There are two main situations. First, ": magnetic current ⑽eaming) applications Compared with a single drive, it can provide: and ^ requires a relatively continuous bandwidth. With N drives: Potentially provides N times the continuous bandwidth of a single drive. Second, the target side = the access time of a given drive (mainly determined by search and rotation delays) is limited to 10 operations per second (I0ps, ι〇⑽SeC ° lld) number. In the best case, an array of N drives can potentially support N times the Ips performance of a single drive. An example is illustrated in Figure 1D. Data Architectures simply attaching a second drive to a system will double the capacity immediately, but may not have a performance improvement. If most of the access is to one or the other of these drives, the performance will be limited to the more commonly used f month b. Dismantling is a well-known technology that distributes data to the available disk drives in 200535609, so that the drive will be required to retrieve the data system, thus allowing Li Tong f p to use the performance of the performance. The characteristics of paper, and heart are the same as those of a typical disk drive (+, and the storage unit of the address is the magnetic sector). —The data of the magnetic field are typically multiples of an exponent of the m group. For this application, the 512-byte size is to be used for the purpose of the month 'instead of limitation. To disassemble ㈣ '-the order or sequence is assigned to these drives and is selected-the disassembly width. —㈣ Disk drives can be identified as 0 and i. The disassembly width may be 4K bytes, which is eight magnetic regions, each of which is 512 bytes. With these choices, the first 4K block of user data ("User 0" in the figure) is the first 4K block stored on drive 0. The third 4κ block of user data (‘‘ user ’,,) is stored in the first 4K block of magnetic“ i ”. The third 4K block of user data is stored in the second ^ 4κ block of the disk 胄 0 and the fourth 4K block of user information is stored in the second 4κ block of the drive. This arrangement is illustrated in Figure 2. This process is repeated, alternating 4κ blocks are stored between the two drives until the end of the drives is reached. If the system has a large number of small accesses to one or two sectors, the two drives can be accessed simultaneously to achieve twice the random access of a single drive. If the system is accessing a relatively large block of data, such as: ⑼ ⑼, the two drive systems can also operate simultaneously to achieve almost twice the continuous performance of a single drive. The result of the disassembly width selection will be discussed in the last zero redundancy (Redundancy) 200535609 As mentioned above, it is a backup copy that can be attached to a drive to keep one of them continuously updated. For the early method here, any write operation is simply copied to a backup drive. One of the main drives is a reliable, hard disk. This technology has been used in the past. Mirroring, or Raid went out until those drives were lost: knowing that you can do it yourself-drive read

Until there is a failure, at its point, the A drive system is selected for reading. Increasing reliability can cause storage costs. Each mirrored disk drive is a technology that must be used for each major disk that exists. ”JL; # ^ h 'for the cost of protection of the two magnetic bang machine array D: the original two of the same size-an additional disk Array. This drive is referred to as "". For this arrangement, redundant drives: each: V 'drive. See Figure 3. The χ £ block of the corresponding 4K block of the disk drive is to receive the other two magnetic failures. The content of the failed disk drive === on any single-drive j 4K block is 荽 斗斗 曾 # The rest The data j hunter should rebuild it. One ... The corresponding 4K block of the redundant disks. An array, which has a data disassembly of the redundant disks across t N disks. The same X0R in the lean material block that is 1 kg is stored in the rest of the redundant disassembly: block: reassemble ^ I / N, of which: the additional cost of several is reduced to a data disk drive: disassemble Solution, the redundant disk drive contains two disk drives :: array. For the corresponding areas of each and μ two lean disk drives (§ 8 200535609, in addition, there is no real disk drive between "data disk drive, disk '~", "As a convenience to indicate the disassembly of the disk drive specified work:" Know that with a high percentage of write Γ the disk drive system tends to become a load. The bottleneck of m and the rotation system tends to balance against read access ^ > π can be the same as the benefit of redundancy: ^ = machine failure, redundancy of the array ..., the performance of the dismantling of the array. However, the reconstruction of a data block of the machine is caused by the ice of the two main chains. Disks ^ Established as a faculty and required disk activities to access the various disk drives and resources of the eight trains ''

Exhale. ^ y The extra processing of Bei Cai for the update of any block of XOR will invalidate the redundant block used for the disassembly, and it will also require the update of the redundant block. As noted above, a system with two drives can provide redundancy by mirroring the other drive, or it can add capacity and provide up to a 2X performance improvement .To maintain—required for copying = Extra: The issue of a disk write command or an additional operation required to distribute or collect shell material disassembled between two drives can be easily handled by the driver software ' It uses one of the disk controllers that does not provide any dedicated array functions. However, 'for systems with three or four drives-redundant arrays' X0R calculations and additional disk activity are due to dedicated hard drives It can have significant benefits, whether with or without regional wisdom. In today's market, the array systems of two disk drives are generally processed by software. Larger array S series = use dedicated disk controller hardware, which can be It is located on the motherboard, insert card, or an external box. 200535609 ik ^^ edundancy Hardware) The industry standard system of mechanical, electrical, and logical interface of the disk drive ^ In the drive system can be provided by Commonly called-Control H or "Transfer benefits" and meet any of the requirements of the interface standard interface and attached to a system whose performance is a problem point of any system, direct memory access (job, D⑽ct Mem〇ry Access) is used by the controller to Transferring disk data between the disk drive and the system memory. As a test for accelerated hardware, consider an array of three data drives plus a redundant disk drive; see Figure 4. Before the drive has expired, accessing a block of data requires only that the target drive is read out and the data is transferred from DMA to memory. It has a magnetic access and 4K bytes The data is transferred to the memory. If the drive fails, access to the same data block will need to read the balance of the disassembly, that is, the same disassembled corresponding block from all other drives. Each of the remaining drives is read and the data is transferred from DM to memory. Even if the three drives can have the same average memory: The 'reading operation for various reasons' will actually be completed in different Shimen, once these two blocks are stored in Missing data for buffer construction. With reference to Figure 35B, an element is read from each included fact that the initial state of the head position and the rotational position are different. Refer to Figure 35A, which shows this non- Synchronized data transmission rounds 0, data 1 and PAR (co-located or redundant disk drives) through the individual heart channel to the corresponding buffer memory. Data 2 disk drive system has failed. —, X〇R operation system can XOR is implemented in order to perform the XOR of the three streams. Streaming, the three components are performed on 10 200535609 XOR in logic 620, and the contents are bad and the paid components are stored in a new block of memory Body 622. Note · A component system can be used in any convenient size of the body and DMA hardware. In this process, I — you need two disks to access: 12K of data is transferred from these disks to memory (use 4K blocks for illustration); W of data is from this memory The read back is calculated for the job; and the data of the byte group is written back to the record. The U-body 622 is used to transfer a total of 28K data into and out of the memory. From the foregoing example, it is observed that the following: "The access to the 4K shell material block only requires 4K data transmission. After the failure, the access system requires a total of 28K bytes of buffer access, That is, the system bus and the seven-fold increase in heart bandwidth load.

2. The XOR calculation will not start until the data block is received from all three disks. A L ^

',,,,'stop. Therefore, the entire XOR process is to increase the total waiting time of the read operation, n L L ^ ^ to search for guards, and to generate a motive to make the buffer memory system as fast as possible. Note ... Although the XOR process may have started with I ^^^^ to take the first two streams, additional bandwidth is still needed to store the intermediate results The final column XOR. I run 3 · After the failure, you will be temporarily suspended. Gu Bei tripled the overhead cost of the drive management. Recognize with Μ_ 料 值 / 0 1

RedundairL ^

Transfers) — typical disks & I, I, and inner buffers, to decouple the poor data transfer rate of the read / write (R / W) header and the honest and honest 4) member " "Dry 1 interface Transfer rate. This internal buffer (the ability of 200535609 to comply with various interface speeds) can be used to enhance redundant operations and significantly reduce hardware clouds and cards. Consider the ATA / ATAPI interface to its original Parallel programming round-in-round-out (pi〇, pr— ⑴ Γ mode. In this mode, the data of a single sixteen-bit block: read from or write to the magnet lift; + + / 、 connector Provided-read bit / 'It is used by the controller / turn or write strobe (DIOR or DIOW). From the above description of redundant hardware, we can know' rebuild a block in a failed drive ' The lean system needs to read the remaining three disk drives, transfer its data to a wide area, and then read the three streams from these buffers to calculate the number of cattle / numbers. Because these disks Machines (although they all operate at the same time) are different from each other, so everyone is surprised to transmit data. ("Internal buffer The application uses the internal buffer of a disk drive as if it is different from one: the buffer memory of the adapter or the host.)-An alternative technology is called synchronous redundant data transmission (S thanks , Pass y: i, r〇nous Redundant Data Transfer). By synchronous redundant data = out command is issued to all three (or n) drives. When: shell material system can be used for each drive The internal buffer is not a buffer that is immediately transmitted to the controller / adapter. 2. However, when reading data from all three drives individually internal buffer, the process is You can start. —The job and the engine are to grab the -th-component from each drive; calculate the tear of the three components and turn the first component of the result into the controller / adapter. This redundant operation is "on_the_fly", as it moves data from the drive to the buffer with 12 200535609 data. If there is a difference from the above-mentioned one hundred stored data in the buffer, operation. The ni-connector must read it out for redundancy. ata / at for PIO mode

} Interface, the component size is a 16-bit character group, that is: 命 命, Rifeng break ^ ητ visibility. Component grabbing is achieved by gating the three drives with the same MR. Make ? The DIOR gating system makes the data transmission "same as above", X〇R is on the face. Under the hardware of the farmer-the JX〇R system will not start until the data of Xing Zitian machine The value is missing at the end of j. A disk has been transferred from the disk to the memory. In the Synchronization (SRDT) plan, it can be used for ^^ ^ Self-fetched data from the disk drive is used for When the internal buffer of the disk drive, the tt tb in 'S σ starts privately. Assume that:, output, and pass are generated by the step data transmission and real-time fetching by the disk drives. The advantages of IT and IT are as follows: h. When the last completed drive is in its cache, from this time, X0R is counted as having read *. ^ And the result is a transmission, which has a long waiting time before the failure of a disk drive. Grab the same transmission from the buffer | ..., M, 4 and save the result to 猞 猞 哭 々 #, t open x〇R, and the waiting time outside the ° ° ° are eliminated.. The total data transmitted to the buffer in this example is required for oil; ^ ... the original ... transmission 4K blocks. Bandwidth. "Frequency is required to support a single drive. • Data from three drives is de-addressed and counted.) DMA content is required. Each drive is a stream as low as a single stream. It is only used for this operation and not required for the original instance. However, this efficient operation depends on the use of equals One of the widths of the disk drive interface stores the component size ("narrow disassembly"), and it is limited by the synchronous transmission caused by applying a common DIOR gate to all disk drives in the array.

In view of the above background discussion, several issues remain. Regarding ρΑτΑ technology '-the controller can access multiple drives at the same time, allowing it to implement redundant calculations in real time for improved RAID performance and significantly reduced hardware complexity. The technology that has been developed (first UDMA, then sata) is a synchronization source, which will not allow a controller to transmit read data synchronously. Secondly, the previous technology used synchronous data transmission and real-time redundancy with the disassembly width of several bytes or words, and combined the drive data into a single Γ stream, which can be a single DMA The channel transfers to and from the buffer fast. Technology is needed to extend the application of synchronous material transmission and real-time redundancy to arrays with a disassembly width of one or more magnetic domains. In addition, it is possible to implement the current stepping technology for the controller to transfer data to memory before any X0R calculation. I: Compared with the possibility of more than one operation without having to transfer all the data to the history first = it is a wasteful buffer for the technology that does not need to read it back for calculation again-a kind of raID drive controller implements disk storage disassembly And redundant operations. Dagger

蟓 m magic port from a separate source (for example: SATA

Port) while multiple drives are connected. A1A logic system provides two or more ㈣ & n lgUrabIe) f material path switches, and dynamic configuration of one attached drive to one or more

14 200535609

Arrays. The data transmission system is subject to regional synchronization by leveraging the SAT. The synchronous transmission train allows instantaneous redundancy. Second, "Good performance" and reduced complexity. Tiredness: Two body ^ Μ)) is to reduce the number of DMA channels or synchronization required = device requirements' and support a variety of narrow = the improvement of some disassembly and update. The eight series are very clear From the detailed description of the following preferred embodiments, the other aspects and advantages of the present invention are well understood [Embodiment] ", ..., the description of the bottom bracket refers to the drawings. Establish synchronous data transmission to and from a magnetic controller to and from the drive 1 'to the disk. For some interface gold protocol Beco transmission is asynchronous. That is, the needles are controlled by individual disk drive electronics at a different time and are completed in one cycle and one of the data of each disk drive is read out = part of it. For example: dismantling the remaining information "ie t Two (:? The availability of material transmission is to enable redundant life (in the real-time) of the missing data of the disk disc Λ * *,) and in the direction of reading this # 可 拉 "(make in —The case of a magnetic disk failure. '"I put an elastic buffer (that is, an example of each interface of the FIF0e drive and controller data), although it can be used as: Slightly two are originated from the storage device of the library instead of one or two, and are selected from each of the drives and their FIF0, =. Any application. For the interface that is subject to the MDA agreement from Xi Magnetic, access to: ：: In the reading of the drive and push it to the right of any one to chat to be close to full, the interface will use

(S 200535609 For the purpose of "suspending" the transfer of data by the institutions provided in the UDMA agreement, FIF〇 will provide _ "almost full (alm0stfuil)," ▲, it is determined that there is enough space left to delete In order to accept the maximum number of words, the "pause" has been determined. Yun: Most of the data in the method described in Wu Guo Patent No. 6, G1 8,778 are synchronously removed from the fIF0. IiA ,

Wait for I. After issuing a read command to all drives, wait until there is data available to transfer to all F

The buffer memory is obtained by using a single address technology device and transferred to the controller using a single address technology. If any of the FIFs becomes "empty", the process will stop until all of them indicate "not empty" again. I consider a disk read operation. Once again, read a few words and read "In the Control of the Early Morning ® Address Leaf Data Fung" r 円 ’円 Chong Yi. The data / segments read from the buffer are advanced to each by using a common strobe, (⑽) = the drive in the array. If either chest becomes "full

Agreement No. 8 ... The process is stopped. On the drive side of FIF0, the interface implementing the UDMA protocol will be proposed from a drive. Although these items are transmitted to them, since each input may start at the same time, it will be asynchronous, and each device will respond to the request of the attached disk drive independently. 1 out. L comes from the use of magnetic tFlF0 or similar memory, which is non-synchronous (sound refers to the drive to generate its data strobe); machine eight ... 16 200535609 = Synchronous redundant data transmission system provides significant advantages over the standard technology used to process simultaneous data transmission requests for disk arrays. Examples are explained in more detail below. Betty 6 illustrates one Disk 10 of the array. The oral reading is used as an illustration rather than a limitation. The disk ... has -data :: 14 'to provide read-out data to its implementation of the UDMA agreement-

16 Similarly, the second drive 20 has a data path U connected to a corresponding @UDMA interface 24, and so on. The drive q item can be displayed for explanation. Each physical drive is attached to a UDMA interface. Each disk drive is lightly connected to the -data input port of the memory via its udma interface. The memory system such as -FIF0 ', although other types of memory systems can be used. For example, the disk drive ^ goes through the UDMA interface 16 to the first-first-in-first-out memory 26 ′ and the disk drive 20 goes through the UDMA interface 24 to the second-in-first-out (FIF). ) Memory 28, and so on. In each case, the udma interface accepts data from the drive and advances the data to FIF for read strobes on the drive. See ... · Write from drive 12 to first-in-first-out (FIF0) memory 26 ... inverted input signal 60, write WR input from drive 20 to first-in-first-out (FIF〇) memory 28 Signal 62, and so on. As mentioned above, this strategy is the opposite of the PI0 mode, where the read strobe is provided to the drive by the controller. If any FIF0 is close to a full condition, the UDMA interface will be "paused" for the method described by the ATCA / AτΑρι specification of NCITS. For this purpose, FIF0 or other 17 200535609 memory systems provide a full (AF, almost full) Chuan 1 ', which is “with sufficient space to remain available for nF0 = denier; expected [According to the interruption, the number of words that can be sent by a disk drive is accepted. The big data is obtained by using something similar to the United States and the United Kingdom, which is described in Patent No. 6, 〇18,77δ: This method is synchronously removed from the compensation. Optionally, after issuing the read order to all the drives, 'there is waiting until resources are available for transmission to all chats, that is: — "Empty (emPty)" condition. This is to say that instead of Fig. 6, it uses the input from each fif0 to-the logical box 4o to generate a lean signal. After all the deletions have the instruction of data, that is, 2 FIFO all have data from its corresponding disk drive, read the data to lose. See the shell material transmission as described later. Each fif0 system has an output path, such as K. 48, in this preferred implementation For example, it is 16 bits wide. Fang: Drive data paths are merged side by side, such as Out of the box 50. In other words, an "Official from the chest to a buffer ^ re)" data path is provided by one office, 11 °. 52, which has an equal to N times m bits and eight middle, N system The number of attached drives and m is the width of the data path from each drive (although it is not necessary that they all have the same width: ―. In the configuration shown in the figure, four drives are used, each of which It has a 16 'lean path, and it is a total of 64 bits to the buffer 52. 44 The transmission of FIF0 and other data is moved by a common read gate, and the common read gate 44 is transmitted to all FIF〇 。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。 FIF .。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。 From which is shown in the figure to the FIF 0. To 18 18 200535609. A mouth P knife. If any of the FIF0, the process will be delayed until the establishment of ^ / ^ (㈣) (empty) ". U refers again to "not empty". Referring to Figure 7, the disk write operation is described. Once again, it is introduced into the controller and each disk drive with a single address counter 7G, and the path # 1 is transported, And 5 贝 comes from the buffer 52 inside the controller. In a preferred embodiment, the value of the buffer slope of 1 § 1 is finally half-duplex (half-duplex) duplex), Jin Sang, Yue Fei) " Hai Temple FIFO and address counter. Each FIFO has a multiplex to cry J to eight ports, depending on the direction of the data transmission. The fishing items come from the buffer The rough space ^ The subgroup of the bee group uses a common return, g 72 to advance to each FIF0, and the common 72 is coupled to the write control input of each fif0, as shown in the figure See: Data Path m 78, 80. 'Write data in this way' is "disassembled, on the array's magnetic machine." If any of the FIFOs becomes "," this is implemented by the logic represented by ... to produce "(any are full), a signal. Ma Man / Yu Gang ’s drive side, implement UDMAM interface ^: 糸 The data from these FIFOs will be transmitted to these magnetic tubes. These transmissions may start at the same time, which will be asynchronous, Since each:: will respond independently to "pause," stop, from its drive; 39 200535609

This modification of UDMA using FIFO to enable synchronous redundant data transfer provides significant advantages over standard techniques for processing the same data transfer requests from a disk array. This standard method requires _ DMA channels per drive, that is, more than one address counter. These DMA channels strive to generate multiple short data by accessing to the buffer, and buist the transmission and reduce the frequency that can be achieved by various technologies. It is determined that the buffered H-bandwidth due to the combination of disk data transmission, host data transmission, and access to redundant data calculations has become a bottleneck for Raid control as a design. As mentioned above, the present invention needs to be targeted; the entire array is only a single channel. . Data stored on a disk array can be protected from loss due to the failure of any single disk drive by providing redundant information. The data stored in a redundant array are the following: it is user data and redundant data that are sufficient to enable reconstruction of all users, assuming that any single drive in the array fails. ^ ^, National Patent No. 6,237,052 B1 reveals that: redundant data meter = can be "on-the-fly" during a synchronous data transmission, real. Synchronous data transfer wheel, "real-time The combination of the three concepts of redundancy and udma adapter (its | 丨 ^ 夂 each disk drive is a FIFO) is to use the hardest hardware to provide a high-performance redundant disk array data path. I. Various kinds of alien and logical operation systems may be used to produce a redundant two-material style. The X0R system will be applied to the current explanation. Referring to Figure 8, Yu Xiangzhi's material flow is shown. This figure illustrates that each of the disk drives 300 of the group is one of the 32 UDMA interfaces connected to the group. Each disk 20 200535609 has a corresponding first-in-first-out memory (34) in the data path, as described above. In the disk write direction, the data block is read from the buffer 35. For example, see data paths 342, 344. Segments of these data blocks are written to each drive. At this point, the logical xOR operation can be performed "instantly" between the corresponding bits of these sectors. x〇R logic 36〇 is arranged to calculate the corresponding bit of each section of the cloth. X〇r, a sequence of redundant sections, which are stored in advance—first in, first out memory 胄 μ, in The UDMA interface is transmitted by UDMA interface 380 to a redundant or co-located drive 39. Therefore, XOR data is stored in synchronization with these data sections. In other words,-the pattern of redundant data is instantaneous, and the 'generating' is a "monitoring (Snoop)" disk write process without adding any delay to it. : Consider Section 9 1] A similar diagram illustrates the data flow in the readout direction. The array's disk drive 300, the corresponding interface 32o, and the πρ0 slow body 340 are shown as before. In the read direction of the disk, the grade is calculated as the data segment read from each data drive and the redundant drive. Therefore, these data sections are input to the ship logic state via path 392 to generate job outputs at 396. If one of these data disk drives (disk drive 3 2 2 in Figure 9) is ρ 生生 4, the result will be 'work evening', then the X0R calculation of 394 will result in a disk that has expired before ^ The ㈣ section of the original sequence of the machine 322 replaces it with the current one from the failed drive: 歹 1 and stores it in the buffer 350 along with other data. This replacement system can be implemented for the reason that the reconstruction of the system has not been delayed until now. The transfer of data from the data enthusiasts, such as 200535609, is more completely explained. The figure 10 in the application is a sequence diagram, and $ 4 According to the present invention, according to the present invention, the FIFO-related signals on the readout of the magnetic disk are directed. As the private north b, Shannai said, as shown in the figure, each disk drive has a different read access time. N A disk drive that has target data in its area buffer determines DMARQ (a difficult request). Then when receiving the CK, it starts its data transfer to just 0. In the drawing , 'Disk 0 happened to be the first to complete and transfer the data and then: Disk 2 and M3 in this order. In this example: Magnetic: The system will write FIF once, and all four positives will be written. The four _ borrowed from the allowable data for the unworked, here is a stand-alone synchronous move out from the real world as ... independent traces to emphasize the fact that in the previous technique, try to solve the problem through redundancy ... θ § Existing data protection system has been solved. The problem is to solve the problem of sigma type. Waiting for two disk reads, == points, the data will be read out from the buffer, x0R; XOR, and return the result. Assume still or six — K 4 access, # 75 For the host and the disk with the buffer, add two: Γ calculates the access system of the purpose of one X0R-the third access, so the requirement, 6x buffering ... 1 wide requirement is 50%. To implement this task, the D-change / write operation system is too slow, so the Japanese department required by the school has been designed for this purpose. The time required to calculate the job is reduced, but the operation system It is still needed. "" Information of the punch-the third time Beishi's new greed is written to the disk immediately. To parity

(S 22 200535609 The writing system of the drive must be postponed M ^ λ ^ j and the brake is not complete. The backward system is accumulated and the parity drive system is targeted for the write operation :: The design of Dado attempts to solve this problem by using RAiD5. Another: all the drives in the array are used by this peer. It is used in the prior art to account for it! The system is designed to transfer data from the host as the data Or go to the middle door of the disk drive I = I 11. Since these transmission gear trains occur at different times, for: 4 = "accumulator plus-), system, a complete magnetic zone or two, two or two It avoids the need for additional buffer accesses, but instead, for a significantly increased complexity. As mentioned above, the present invention does not require its buffer for XOR calculations to be compared with 50. Device bandwidth, _X, buffer space to store several lean materials, or /, continuous read / modify / write operations on the engine, or its storage from the computer. In the embodiment, _magnetic ^ _; j is used to read Jie. Board. For example, it can also be implemented as :: ... A host bus adapter (HBA, Host us apter) is used to interface to a ρα host bus. The application of the Data Path Switch stems from the previous "synchronous redundant data transmission," the patent reveals :: :: Disassembly method 'where two or four data drives are simply interleaved. The “array”, described below (also in = a poor material path switch, ′), incorporates new features and methods, Huang Yun = disassembled on three data drives. It is important that this system has only four drives (total). Other uses and advantages are described later. The physical drive port, mapping register, The following description of the data path with 23 200535609 should be noted that other implementations or embodiments may have other numbers of physical ports, such as: 8, 1, 2, 16, etc. These changes are based on this Within the scope of the invention. The disclosed array opening relationship includes its features that are conducive to supporting RAID5. RAID5 is optimized for small random access, and RAIDXL is optimized for large sequential access. RAID5 Performance is usually measured at 10 Operations per second (IOPS) and relative to Megabytes per second (MBPS). These characteristics use the existing Aimed at the "real-time X0R" X0R hardware and a new single-sector buffer, accumulator, and proper sequencing to achieve RAID5 functionality. The main RAID5 functions supported are: completely disassembled and interpreted as invalid Drive FULL STRIPE READ WITH FAILED DRIVE), READ FROM FAILED DRIVE, READ FROM FAILED DRIVE, FULL STRIPE WRITE, and PARTIAL STRIPE UPDATE. Array open relationship implementation data path The configuration of the array switch defines an array composed of one or more disk drives. The array of a single disk drive is a JBOD (Just a bunch of drives, only a group of drives). For RAID0, RAID1, RAIDXL, and RAID5 involve multiple drives. In a given instance, the array opening relationship can be any one of these functions and can support all of them at the same time. The following definitions apply to this Paragraphs and associated diagrams: Array (Airay) ·· A subset of the drives attached to a controller 24 200535609. The listed data drives are LI, L2, and L3. Combined. Array systems can A logical drive consisting of one to four drives:-arrays are numbered starting with zero. The logical drive is L0, with a single redundant interface to a SATA0, co-located drive ( Parity Drive): --Array I may have or parity drives. Parity disk drive system will paR. SATA ports (SATA Port) ·· a SATA port system provides a serial ATA disk drive and meet the specifications required .SATA port system

SATA1, SATA2, and SATA3. Physical Drive ... A drive will get its identification from SATA 4 to its attached. The physical drive will be drive ° B〇 (Dnve 〇), drive i (Ddve υ, drive 2⑺, and drive 3 (Drive 3). "Sector ·· Magnet Disk data of the smallest addressable block. For the purpose of this case, a magnetic block will be 5 12 bytes. LB A: The magnetic block of the drive is composed of a logical block address (lb a, It is identified by ck Address). LBA is designated as the maximum number of capacity of the drive from zero to as high as it needs to be addressed. The data disk drive in the array. A disassembly width is selected. The capacity of each logical drive is considered as a group of disassembly widths. The capacity of the delta array is mapped to its logical disk. Machine order, the first of these blocks of each logical drive of the array, then to the second block, etc. For 1 ^ 100 or RAID5, the disassembly width will be a "two" Power of two "number of magnetic domains. For raidxl, the 25 25 35 35 609 resolution will be a DWC) RD. RAID1: in this mode With a single logical drive and a pAR drive, it holds a mirror image of one of the contents of the logical drive. This is why Wu's e-drive system looks like a host system. Is a single drive.

RAIDXL • This mapping uses DWORD interleave, regardless of the presence or absence of parity drives. In this mode, the drives of the array including co-located drives are presented as a single drive to the host system. " Data Map · Notes. In Figures 15A to 17: The drive numbers shown are arrays of "logical," drives. All values are the serial numbers of the blocks of disassembly width. [+] Indicates the listed X〇R of the block. Marks (η), (n + 1), (n + 2) are those that include some of these blocks. • • • It refers to n relative sections of logical drive data. Data map, line-indicates the axis. Crossing over] shows that for two, two, and, five magical tools; one four-drive array across one of these available drives F material points: solution ^ solution is simple ^ selection, that is: I6K byte user data. For the size of the first 16K bytes, a drive array, times ... user data Top =: Γ drive ... Its user data (called block 2) is performed on the disk drive three I16K ... The resource_ is block 3) is performed on the disk r; and :: ;: 26 200535609 Repeated until the process of these drives is similar. ， 'Two filled. For three and four data magnetic disc 15B figure shows the size of data mapping for RAID1, One of the first two lines by two drives mapped. Doubts whether dismantling system use the term "mirror Cui only beryllium. (The applicant's RAIDXL and does not have 7 maps are displayed for RAIDXL and the right "shows how much data is available from the B map. The user data is disassembled into a rigid frame. , So that the model system can be provided at 16… ', bits. (Note: I. ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, to to, s, the data of, the 1st storage controller ... —), A Jujube machine array, a user, and a magnetic area are stored in the pair of magnetic tablet machines < f & .π, area. The magnetic disk drive 0 has the user's first material / plant first Two magnetic eight 3 1 text used 耆 贝 耆 magnetic zone 〇 盥 ~ As indicated by the label t ", 1 [0]. The disk has an odd number group of the magnetic zone ... if the label =: 者 = 石g She always said 丄, iJ is only sunrise. The co-located disk system has user data η 1Γ Ί Beco's first XOR of a magnetic zone, as indicated by the standard ° 1 +]. By it Bu; 0 〇 〇 佶 秸 This is the way, "8,9,10," [3] means that the user contributes the magnetic field 8, U every fourth block. Array switch (Array Switch): Array Open Relationship There are four sets of registers, one for each SATA I. These registers are used to define individual disk drives or arrays. Each temporary storage 1§ has a mapping (Mappmg), data group length (such as Canada Post), Fast Read, and Command (C0mmand) registers. 芩 Read 18 pictures and detailed configurations detailed below. In a preferred embodiment, the host software driver Load these mapping registers and configure the array of 27 200535609 switches. Thereafter, by way of illustration and not limitation, the system shown in the figure performs the RAID function shown in Figures 1 and 12 based on the current configuration. As described later, the woodwork and the improved mapping field are provided with a bit block system for each real estate and rough drought to indicate that the corresponding physical port / ,, '六 口 口 ~ 私 ~ μ 土Is the last money used for a specific temporary storage array ... If the drive is used to determine whether it is a parity drive or a data drive. 1. The situation will indicate the logical magnetic Drive ID: The effect of the drive. And whether the drive is missing, The length of the data set is essentially the length of the data set of the number of logical data disk drives that are to be continuously transmitted. The quick readout flag indicates that: a redundant array with no magnetic effect is in the same position. Whether the data is read out for -reading = check. If not, the readout is due to reduced spin waiting = faster, ie: "quick readout". It will be loaded for the command register. One of the basic actions of this array switch is defined as follows.) JBOD (Just a bunch of drives A heat recognition 丄 m% 18A® es, only-group of drives) Participate in this array switch Preset content, which follows the power-on reset. Each SATA port / physical disk drive is an independent stand-alone array, which is applied to the corresponding DMA channel of the host interface to be proud of :: Passing round. The software drive will send a command to the SATA port. When requested by SAS, the array connection will use a DMA channel to transmit a single magnetic 28 200535609 packet between the SATA and the buffer of the PCI bus. RAIDl ·· Refer to Figure 18B. This configuration always involves two drives', namely, one co-located drive. Gu Xuanying's grandson Yu Hengyi is true ... What the shooting system shows is ... The logical drive 〇 is attached / the cockroach cock 0 and the parity drive is attached to the physical cock. Parity maintains an exact copy of one of the lean drives. For RAID1, the array of a ^ 'machine is, for a disk, Λ ▲ is an earlier disk drive. Yu-disk drive write, the array switch logic caused the write command :, the machine, and the same data is written = / solid disk command is only sent to the resource manager sincerely secrets 佚 drink out 'Gan Chu 4 Drive and only that data drive is attached. For ten- "slow read,", the command is transmitted to the two drives are both access, and parity (PAR, parlty) = content is checked against the content of the data drive. The second line of the = line indicates that "埴! 士士 及] and the color bars are used to indicate that the memory system is unavailable. 'Since this port system has been used as an array defined by port 0, ... Two RAID1 · Quick Readout: Refer to Figure i8c. This configuration is similar to ^ e,. M, and. A disk drive is involved and its content is always the same. It is the second one. In the fourth temporary storage = wide array system definition — —…. ° group L0 is in port 3. As mentioned above, “quick readout” is only for accessing data drives. & — a different disk drive, ”etc. ::: human array” 0 access at the same time. For writing different parts of the rf data set, only one of the waiting drives is a two-body drive. The system must ensure that this will involve writing and attempting to read at the same time. Han Zhijun 29 200535609 In a RAiDXL, the data is disassembled into a DW〇RD wide and span data drive. The first __ is stored in the first series of cars: machine; the next RD〇RD is stored in the next logical drive, and so on, until each logical drive has been received by the Russian department a DWORD. The bit method of DW0RDS on the disc drive ⑽㈣e) χ two different DWORD #stored in n y ^ 50th germanium is stored in the position drive (if it exists). This is repeated for the entire disk. —The given volume makes each data drive in a row. In order to ... A Beco system will arrange the blades in the array to perform access to a given magnetic area, and all drives in the array must be accessed. The small addressable data block formed by the magnetic tablet machine of the Haihe array is also a magnetic zone, so any entity is at least one magnetic disk of the disc drive. Long accounts and * make up the minimum transmission length equal to N magnetic zones, where N is the number of data drives. From 丄:, all the sections are stored in each-疋 disassembly, the same relative position of "位,:" required to access these sections is the same for any access. This allows the array to be propagated to all drives, making access possible β drives. For a given drive, the difference: the drive system will have -error. Any, the block removal system must be resolved by accessing to the 1 disk drive. See table 8 if the specific magnetic disk with the same position exists,… and the f # # # & The new calculation of this parity disk is compared with the data from the heavy disk drive of the door / out access ... class of the material.-Error refers to 30 200535609-is not matched. Material # is not added Fetch time, but access to a disk drive increases the average waiting time for access.: The option of just parity data can be quickly read by "㈣, but refused, Pakistan's W, parity drives are always accessed for writing. The benefits of co-located drives. The problem is that even if one of these drives fails, the array is allowed to be insignificant ... "• The failure of H-bit drives is R Na The alpha material is purely with the same position-the new corpse ... one or more data drives on the right are invalid, and the array opening relationship is == indicates the position of the failed drive. The indicated magnetic disks were not received Any dissemination order or participation in any data transmission and writing and what is the cause of the ^^ 'Parallel drive system will be accessed by reading Shan // Ma Ma. Yu Yi ^ ^ ^ Μ m " $ All remaining data disk drives: will be calculated. The result of this calculation is equivalent to the data stored in the failed disk drive. The management of the data input data of the drive is discarded based on all writes. P causes the data software software program for the failed drive to receive from the operating system: The edge Γ extends to one or two magnetic regions at each end, if necessary to achieve-; angle + boundary. Needle oblique to bite ~ ^ magnetic The number of disc players = a command system is established, by dividing the LBA address obtained by the number of data into the thinking. The extended command system will be flat t? X king-Lu Dingzhan and ^ ^ knife.〗 There are two data drives on the right. The WBA's LBA and Lai will be only the LBA and the account 31 200535609. The only open relationship that stores this data will be to merge the data from the array ... 1/3. Array /, shell material to a single stream, used to return to the user's buffer of the transfer wheel. The driving software program must handle j shi ", f male tools have extended your field to … Is about to transfer the total request calculation program is to create-scatter list (scatter ten 's out, ask for any of the front-end or back-end of the data: the daily quote attached to the data is directly transmitted. Discard the buffer for a write The request will need to be targeted at the "" ,， 方; the additional magnetic field system at each end and the Wending Town tile β ί read_repair / repair / modify / 耷 人 π " write operation. For In a read / 1U change / write operation, the drive includes to -㈣ ... The flying system disassembles the H buy target, and its (gath er), the direct pick-up person will create a collection of data for the user and the user. The extended magnetic area of the resolver (such as XOR operation can be "real-time" real XOR accumulator FIFO hardware, such as Examine the published patents that are not shown in Figure 6 and Figure 4. 胄 For the whole explanation, please refer to the RAIDXL-2 Drives-No Parity (RAIDXL.-, parity) listed in the previous section. Refer to Figure 1 9A for a drive- What is not shown in the figure is: logical disk 胄 〇 is in the compilation drive 1 is in the physical store a, and

-... h The smaller length of the transmission wheel is the magnetic disk drive for the parent of ㈣A--a magnetic ... The number is transmitted to both. The order from the “two-in-ones” to the shell body 0—the transmission of all-in-one shells—please prepare for the transmission of data. The data is used by its counterparts until a single dma α, The array switch temporarily passes the MA channel. The gray stripes are "1 32 200535609 Body Port 1 is used by the current array.

RAIDXL-3 Drives-NG Parity (RAI parity): Refer to Figure 19B—a disk drive-no disk is: Logical drive G is logically connected to the physical port ", and is connected to logical port 2 and logical drive 2 The total is ... "The second transmission request is ignored until all three are: = data. The data is transmitted by using the DMA channel corresponding to the array opening. Six Mody y mouths, and one early, two, do not color 11 卞 pattern points out: the physical port 1 bathroom 2 is used by the array of free rain. The synchronous and redundant input of the reference documents shows a simple disassembly. The 1st patent of the stomach infusion is disclosed, and the number of the disk drives is one of two # t (power) and comes from the data of these magnetic songs. In the present invention, chopping: early interleaving between the two Seoul to re-establish the use of Chinese-Zhongji, Zier 1 is introduced to extend the concept to the array, /, the number of disk drives can be different from the two Concept, test, one of the powers. It is an array with two data drives. The data path to the shell disk drive port is DW0 / to the surface data path is QW〇R host system refers to a magnetic zone, which, make :: 0; :: access to each of the data stealing drive Single magnetic domain. — Distributed among the three

Ten districts. The hard system will read out synchronously—A comes from the physical disk drive data port of the temple—: Ren Qi's six DW〇RDS. Obtained from the _ mouth to 'to generate a total of RDS system to send two DW〇㈣ two packs M-> 戌 system, and one to keep in a temporary crying brother one out, hard The system will hold a total of four bribes, which will be sent to the host in two cycles. The first cycle is dw〇rd from the scratchpad and DWORDs from the second read. The first and next cycle is the balance of DWORDS from the second read. This process is repeated for the balancing of these magnetic zones. 'Parity'. Refer to Figure 19C. The picture shows: logical drive 0 is on physical port 0, drive 1 is on physical port, logical drive 2 is on physical port 2, and ^ series Magnetic «3 is for physical port 3. The minimum input length is one sector for each data drive that is four: Let the system spread to all of them.埠 The fate of each port is fate ^ Straight from the ancient times, the enthusiasm of the heart is not to be ignored and the array gate ... The fixed port is ready to transmit data. The data uses a single DMa corresponding to the #guan Yue memory device group to point out: the physical port m3A " grey stripes

# 3 is used by the array of rules. IDXL-3 Drives-Parity (RAmxL bit) ·· Refer to Figure 19D. A drive · The same is shown in the picture system. · Logical drive. Drive 1 is physical! , And the parity drive is two = 0, and the logically small values y and 戍 are in the continuous port 2. # J's pass length is for a total of two rounds of 2 each. The command to write to the actual production n + drive ^ we drive the drive to a magnetic king shell body port 0 is propagated to 1 _ or "slow read, 'write to entity Tan 0; = value is for Write three bees. For "quick readout", the command ": spread to all take 4: Γ from Ren-cockroach data transfer request is ignored and = there are f 'i all saved DMA channels while 34 200535609 pass. The gray stripes indicate that: Entity 珲] and Entity Chun 2 are used by the columns. Μ Ι RAIDXL-4 Drives_parity (raidxl_ four disk locations). Refer to Figure 19E 'the same as the figure shows: logical drive 0 is on physical port 0, and Luo drive 1 is on physical port 1. The logical drive 2 is on the physical port ^ The same as the drive is on the physical port 3. The minimum transmission length is one sector for each shell drive with a total of ^. For the command system of writing 0-speed Γ ′ to write to the physical port G and propagate it to all three H fast readouts ”, the command system is propagated to only these data ports. Data transfer requests are ignored until all access ports are ready = input. Assets are transmitted using corresponding difficult channels. = ^ Points out: Actual! 'Entity 璋 2, and the array of solid bees. A 田 目 刖 disk! 1AIDXL 3 DrivesParity-Drive Faiied (raidxl-three "parity-drive failure) ·· Refer to Figure 2A Figure 1_This example 'corresponds to the figure field of physical port 0 The system has been modified to take the disk drive as invalid. This disk system has been stored to two: then the logical drive of the array will no longer be for the target system ... Configured in this figure, the physical port 0 ',,,, For read or write, the parity drive will transfer the A t data by = 1. At the time of purchase, the weight of the failed drive ::, the rest of the array ’s Data drive: Data: L. At: At the time of writing, 'It is written to all ㈣ R Zhi Ye Yi and stored in the same drive, even if (§ 35 200535609 疋 it wants to write to invalid The data of the drive is discarded. RAIDXL-4 Drives-Parity.Drive Failed: See Figure 2B in this example, which corresponds to the physical entity. The block system in the figure has been modified to: .., member is not ... Hai drive is invalid. This drive has stored the logical disk for the current array Data! Due to the configuration of this figure, the physical port i will no longer be accessed for reading or writing, while the co-located disk drive will be accessed for reading or writing. When a read The data for the failed drive is re-created by calculating the xOR of the data from all the remaining drives in the array. At the time of a write, it is written to all logic: stone «The grade of data is calculated and stored on the same drive, even if you do not want to discard the data written to the failed drive. RAID0 (General information is wide, in RAJD0 'The disassembly width is a" multiplier of two " Square (p 0 er 0 f tW 'number of magnetic zones. For example, regarding the disassembly width of one of the 32 magnetic zones: take the user data of the first 32 magnetic zones and store them on the logical drive 〇: The 32 sectors are in logical drive 1, and so on, until 32 = are already stored in each logical data drive. This process is repeated for the capacity of the disc. The function of RAm〇 uses s "A璋 / disk drive is used for the overall implementation of ΓZΓ in software. This system is without any hardware support of 222 1ΓD 0 Upon receiving the access from the operating system, the software driver will decompose the request into a sequence of storage == disk: this includes: finding only the user's decision processing, and reporting completion. 36 200535609 …… general information): RAID5 is on one side and the same drive is 1 = 2 is different from ™ 0. First, there is a ^ U stored on the lean disk-a yuan method grades are calculated and stored in 1 Door ~ 疋 The dismantling position of the information is-, the section dismantles; the corresponding disassembly of the co-located disk drive. Di Yi Gui Temple-Bei Lian and the co-located magnet are in the disassembly, and its square is _ ^ 八; ^ 之The break-to-body alignment is a rotating disk player. —RAID5 all the magnetic machines in the array) #phase ππ 1 for normal read access (which does not involve co-located disks that are the same as _ RAID 0 array read access rod software driver must be allowed, nothing The n, 11 t series to the rotation of the solid-state disk drive is disabling when disassembling a disk drive-the disk write operation or the reading function: the disk drive access is more complicated. The system includes The hardware work μ helps the software driver program for this work. Magnetic Zen Γ RAIDXL—- continuous stream of user data is interleaved between data / '/. For this reason, it is expected that any access is convenient for early use. One DMA channel, and the Phenix array, P, I transfers Beco as early as a user buffer and solution: In RAID5, store in-a given data drive to remove the 'sector system' The solution that can be multiple kilobytes (kilobytes) requires data to be transmitted between the data buffer and the disk drive :: ,, =: The data buffer is located in multiple kilobytes. For Here: '-DMA channel is used for each data drive. These sectors are transmitted sequentially and share a single DMA channels, but this is restrictive and means a large number of buffers to perform job calculations. Furthermore, for the JBOD function, each disk drive already has a -βΜΑ channel. The register of the switch must be The program is planned to be basically 37 200535609 action for each ㈣. This is the qi & m μ j for several years; rotation between dismantling and also for borrowing its normal designation one: ^ (RAir! ^ I? S_FU, 1 8Η Ηββ RCad 21A drive · Completely disassembled due to failure of the drive): Refer to the example,-disassembly system access, where the logic ... ΑΑΑ 〇, and the co-located machine 1 is assigned to SATA〗 / 3 logical disk SATA, but this drive system is no longer valid. Because: material = replies to two drives, the minimum transmission length is still ^ even if one of the data drives-it is already Invalidation. — Dma through ST II for each logical disk drive program planning, including its disk, ,,,, and channel 1 of failure. Must happen here is the disassembly of storage in disk 0 The sector will be transferred to the buffer indicated by DMA channel 0. Stored on disk G's section must work for the section stored on the same drive, in order to re-establish that it is already or will be stored on the drive! Data. Results must be sent to Lai channel] Buffer. Although these access systems are not related to RAID〇, they must be synchronized to achieve XOR calculation. The array open relationship has a single sector buffer, and the output of X0R logic can be stored to it. Command It is written to the actual material 0 and will be propagated to 4 1 and 3. The array relationship will wait until all the access drives are ready to transfer at least one sector. Then, the transfer will be from the first non-sector. A sector of the SATA port of the failed logical data drive to the host that uses the drive's DMA channel. x〇R logic monitors this transmission and

(S 38 200535609 captures a copy of the magnetic zone in its x〇r buffer. In order, one of the balances of each of the non-failed data disk drives such as σ Hai will be transmitted, 'progressive' Each new sector and the current content of the buffer < x〇r. Next, a sector from the co-located drive will be taken, and it will be w with the content of the buffer, and the result will be transferred to its use The host of the DMA channel of the failed disk drive. At this point, one magnetic zone from each active port and one magnetic zone to each host have been obtained ^ ㈣ The main money buffer benefits. The process is repeated until all two ',', ',,! Transmission. If the drive must extend the user request to generate one: full (㈣ dismantling request, the attached sector can be discarded using the spread (s,) machine' as In the case of RAIDXL. RAID5-3 Drives_Read removes the failed drive with ㈣a · three magnetic tablets) ·· Refer to Figure 2ΐβ, / Example of the disassembly of a segment stored in the failed drive Same as :: In the previous example, logical drive 0 is assigned to SATA〇, parity disk This is the DMA channel assigned to the SATA 3 0 machine only. The effective disk is the same as before, and this will be the only data transmitted. Examples of this disk, including the same drive All remaining commands of j are propagated to all access drives * and until all access drives request a request, h will wait for the tablet: the receiver will synchronize the transfer to each store. Take the drive: The data of the machine is transmitted synchronously, the data of this data: η magnetic and the result is stored in the DMA channel, which is 4 calculations, "out of the host buffer. This process 39 200535609 is repeated until all the data has been transmitted. RAID5-4 Drives-Ful, Stripe Read With Drive Failed (RAID5-four drives-completely disassembled to interpret the video as "drive failure" see figure za in this example,-disassembly system access, where the logic Drive 0 means 疋: SA: 0, logical drive 1 is designated as “TA,” and the co-located drive is designated as S SATA 3. Logical drive 2 is designated as “η 2, = This: The disc player system has become invalid. Since the data format is disassembled from the three magnetic ones, the transmission length is still two. For three 'even if the drive of the data drive :: Economic failure:-0ΜΑ channel must be programmed for each of these logical drives, including the drive is a failed channel 2 What must happen here is: stored on the drive 0 will be transferred to the DMA channel i refers to ... and it is Λ. Τ out of it, You Chong is. The segment stored on the logical magnetic 隹 'Λ 1 must be XOR with the segments stored on the same drive to re-establish it as data already stored on the drive 2. The result must be sent to the buffer indicated by DM A channel 2. 'Although these accesses are not related to RAID 0, they must be synchronized to achieve XOR calculations. The array open relationship has a single magnetic zone buffer to which the job can be stored. The command is written to physical port 0 and will be opened for, ^, 1, and the 3 ° array will wait until all :: drives are ready to transfer at least one sector. Next, one sector from the SATA port of the non-effect logical data drive will be transferred to the host, the drive's DMA channel. The ship logic system monitors and captures a copy of the magnetic area in its buffer, which is sequentially transferred to 200535609. It comes from these non-failure areas, and each new magnetic progression is advanced. It will obtain the same inner valley. Host that performs XOR, DMA channel. One of each of the balanced disk drives is x0R and the current content of the buffer. Connect a magnetic area of the disk drive, send it to the buffer, and send the result to its failed drive. The π method of using the failed disk drive is from a magnetic area of each active port to each host. This process is repeated until all ^ 8 · β… ^ machines must make Xiao Zhe request to produce-Wu Wang Wang Jiaojiao Wuqiu request, the attached magnetron is like RMDXL. Use the dissemination mechanism and discard it. RAIDS_4 Drives_Read Faiied (drive-failed drive): Refer to the 21DS four disks. In this example, store the failed drive. As in Yu Yang 1 ’s example of a sector system that was dismantled from a card, the logical drive ## γ + logical drive i «Set to SATA 丨 4 series & ^ SATA0, SATA 3 . Only for w Ab and the parity drive is assigned to channel 2) is the program plan, ^ (in this example is the previous example, including the only data transmitted. As the machine access. Command Pass on all the remaining disks of ^ j and wait until all the disc players in f / spoon. Array open relationship

Data connection: For: the disc player requests a data transfer wheel. -One magnetic area Ding Shiyou will transfer the data to each L drive simultaneously. Ticket machine. From all calculations and processing results = tearing of this data, "real time" and ,. If it is stored in the DMA channel, the process is repeated% j host buffer. This spirit and until all information is passed on. (S: 41 200535609

RAID5-3 DriVes_F machine-completely disassemble and write) ·· Moxibustion door ... Training Write (RAID5_three disks ^ 翏 Read the 22A figure test 2nd figure This example, logical drive 0 is assigned to SATA 〇And logical drive "means

0 〇 to SATA 1. The parity drive is based on SATA 2. Injury due to poor material format. ,,,, and undressing two disk drives, the minimum transmission length is known. D Μ A channel you 7 / must be written to SATA 〇 for each logical drive and will be transmitted to Μ 〇, ΜΑΒ and SATA 2. Array open relationship

J Zhu will wait 4 until all functions are SATA 璋 ready to receive data. The technology ^ ^ iA, the player 'will transfer a sector to the logical drive SATA 0 of the SATA 0' using the buffer indicated by the DMA channel 0. This transmission is monitored by the array switch, and a copy of the magnetic field is captured in the XOR buffer. Each of the 4 ^ 〇 magnetic domains is sequentially transmitted for the remaining data disks, and each of them is sequentially transmitted. With the passage of each magnetic field, the XOR system progresses slowly and slowly. With the final magnetic sector stored on its data drive, XQR ancient times.卞 t and 纟 σ are stored in the same disk drive at the same time. At this point, one magnetic area has been taken from each host buffer, and one area is written to the data Drives and co-located drives. This process is repeated until all data is written. As in ㈣ 胤,-user-level access may involve less than a complete disassembly. The user request can be extended to the dismantling boundary. These extensions will require read " multiple change, write operations, as previously described. RAID5-4 Drives_Full StHpe WHte (RAm5 four drives-complete disassembly write): Refer to Figure 22B and refer to Figure 22 again. In this example, the logical drive 〇 is assigned to rs 42 200535609 SATA 〇, logical Drive 1 2 is assigned to SATA 2. π SATA and logical drive ^,. Co-located drive system, SATA 3. Because of the information ^ $ sticks, the length of the small round is three. 1) The eight-channel Laos "must be programmed for each $, δτδ λ Xinger's disease. Φ order is written to SATA 〇 and will be transmitted 传播 ΔΤΛ, W 雀 to SATA 0, SATA 1, SATA 2, and bATA 3. The array is open for supply / ice, and waits until all the functioning SATA ports are prepared for receiving data.

One sector will be transferred to logical drive 0 of SATA 0, using the buffer pointed out by DMa, storing A 妗 ㈣ MA channel 0. One copy of this magnetic field monitored by this pass was captured by the office workers: the "magnetic field system" was then transmitted sequentially for each of the remaining data disks. As each magnetic field passes, the grade is accumulated, and the final magnetic field system stores the tablet. The XOR calculates the total amount of the data. The result is stored in the same drive simultaneously. At this point, the solid magnetic area is a magnetic disk that has been taken from each host buffer, and one magnetic area has been written to each data disk drive. This process is repeated until all data is written. For example, ‘one user ’s order sound: It ’s possible that the month & involves less than — complete disassembly. The user request may be extended to the dismantling boundary. These extensions will require read / modify / write operations, as previously described. RAID5_Partiin StHpe π · Partially disassembled) ·· Refer to Figure 22C. Referring again to Figure 1 of RAID5, a user may write to an earlier volume. The parity drive may be updated without the need for a complete disassembly-read / modify / write. Only the target data drive and parity drive 43 200535609 The shell material system will change. This is to keep other disk drives available for simultaneous read access or other pairs of disk drives available for simultaneous partial disassembly and update. With the update, the parity drive must contain its entire disassembly XOR 'including the new data but it already has the entire disassembly XOR with the current data. The traditional RAm5 approach to this problem is to read out the data segment it will replace and read out the current parity data. The two are XORed to provide a parity between the disassembly and the target data area removed from the result. The new data segment is then written to the array, and: the first 4 different results are performed XOR to generate an update and disassembly sequence of its packaging segment. This system is written to a co-located hard drive. The switch hardware includes features designed to facilitate this process. As mentioned above, the partial disassembly update involves only 4 drives and a co-located drive. As shown in the figure and the example of the disk attached with a life saver ΑΤΛ η, the lean disk drive is considered to be connected to SATA 〇-the logical disk 胄 5 Q △ lower Λ Q 问 I the disk drive system is designated to SATA 3 . Since there is only one profile —. Included in the eighth round, take the length of the small round and write it to SATA0 and spread it to SATAmata. During the entry phase, the array waits until the port is ready to transmit data. Following ΑΤΑ, the shell material is followed by X0R as the host buffer indicated by the channel. The result of the formula of ten f 3 is for the Λ contention * fp ', k ί, and the stylistic driver r each other. …, There will be updates for the parity data of those who have been removed. The value reached here is also the current magnetic track, which has become a traditional RAID5 hard mountain " female., Two read operations and the first x〇R ^ 丨 more The leaf that was changed / written was opened earlier. Refer to item 22D. During the writing phase, the filial piety driver program 44 200535609 is planned with the eight channels 0 as the address of the buffer that holds the new data and DMA channel 3 has the address of the parity counter that it has just completed. These commands are written to, and Friend King α 铒 珲 ϋ and transmitted to SATA 0 and SATA 3. The array will wait to receive data . Then, the following will be used: =, the drive will use its DMA channel to transfer a sector to the sATA port of the drive.

Reproduction is captured in the buffer cry. Then ::: heart · 1 and the share of this sector, the balancer will use the SATA DMA channel of the same drive to transfer a sector. As the data is transmitted, the relationship between the data and the buffer will be calculated, and the data will be sent to the co-located drive's SAT ° " read from each channel. : Each magnetic area is a SAA port. This vortex _ raft tongue, — ^ G ,,, and 冩 冩 至 to lose. It is hard to be private until all the information is passed. The previous narrative shows the mapping of its organization in logical order. What is presented now is: a comparative 9 σσ. The real solution is to map the register for the organization in the second edition. At each temporary m there will be specific stops. The board and machine at the stop (or pointed out that the logical disk created by ΓΓ is not used in the array). π · Mapping register and data path switch logic register. Typical raids for a small computer system include: to-the host system-interface, and to a disk array 2: system package: mention = map system : A simplified block diagram of the disk array controller 10, interactive gray-host interface 12 of one of the host bus 12, and 45 200535609 for interacting with multiple attached disk drives-the controller is better For inclusion, the drive interface 22. A control processor body 18 is used to move to the host ^^, the human-buffered memory, the copying side, the host bus and magnetic storage. Temporary materials between the management-the physical port system needs to be attached-a large number of storage machines) to a system. Although some mediators: (block such as multiple devices on one disk, the physical port is tilted, and the monthly data transfer to n R: becomes a — bottleneck. For this reason,-questionable month & The RAID twisting system can be used for each large-to-physical port, as expected in Figure 24A, and has the corresponding content of the register 24. The figure also shows a mapping temporary performance benefit two = 1 to further explain these disk drives. For example, _ shell material is disassembled in the array ° _ people read from four disks 嫱--owed "丨 generated for a single drive "The example shown in Figure 24A of the disc player's material, since ° good. For _ ten people who have arrived from the four disk drives from the poor material system to logical drive order,, Xin as, Fourteen bits of information are oblique and sent to the buffer (as shown in Figure 23, Section-Dismantling, 卽 .JL A av, Friends 0 °. 18). User data is a decorative solution, namely To distribute one at a time—area 1 ^ for example, a 16-bit word) across an array in a pre-ordered sequence " starts at logical drive # 0 and advances ^ Back to logical drive # 11-1,1, the preparation of the drives attached to the array "11" ... This disassembly sequence is repeated to make the user kth area Duan Weisi · Ying Yu logical drive (k mod this way, using logical drive number to reflect the dismantling order. In the figure, the stacking of the four "logical ports" is only to indicate _ disassembly One of the four sections of an ordered set. Each "logical port" corresponds to the

46 200535609 The solution is a single segment, and the whole heap 4 corresponds to the four segments of an ordered set. The 100 MBPS transfer rate from each drive becomes a 4⑽S transfer rate to the buffer. The concept of the dotted square% represents that it is one of the data path switches described in Wuhou. The data path switch 26 is a dynamically configurable data path between the logical 2 material port and the physical data port. The 24A diagram is only a conceptual diagram, which is shown as a direct connection between the logic! ::. For real applications, the available :: The number of shells will be greater than the logical number of materials left as "popular spare parts (SP⑽)," or its security group for its independent access, only the system can be funded. , ... Array. Figure 24b shows the four logical shell ports (logical port to logical port # 3) to the possible designation of _ individual two-body shells with large arrows that can be equal to (real object to physical object) 30 In the case of port cattle, 24B ISI refers to only the designation of # 2. Temporary storage == the corresponding content of register 24. Here, since the second son-the second brother on the right The field is from 4 which indicates the value of the actual port illusion "2", the mountain k series port illusion, and it contains the path switch 26, which is indicated by the arrow 30. After the data path. ^ 4 Designation, such as the complete description in the second picture series display-in the species, each drive system is designated to 51, one instance, the other, that is, physical objects and physical objects. It's used to itch the two physical data One of the 64-bit blocks, each of which is used as a buffer—for the first read, # Fan's drive system must read-sub-out pair The data of logical rectification and magic are in ν. The shells are respectively obtained from 47 200535609 perforated material. In the second reading, the logical volumes # 2 and # 3 are from the perforated material. This The other operations are shown by the row "same" mapping register to the actual material # 1 and # 2, and the flat 24D picture shows an array of one connected to the entity. For this configuration, a "The configuration of the machine 1¾ is obtained by reading the same physical port for four times. One feature of the synchronous output described in US Patent No. M18,778 is that it allows redundant data. "Instant ,,,, processing = as described in Wu Guo Patent No. 6,237, G52. Figure 25A shows an array of four drives and adds logical space 36 to calculate a redundant pattern. 'It is stored in the disk arithmetic and logic operation system attached to physical port # 4. It may be used to generate seed logic. The advantage of the logic of the corresponding bits of the data is that the ship's operating system is not ㈣〇0. Due to the use of X0R, the fifth disk drive system is redundant, _m 心 _) , „Machine / is a 16-bit wide bus grade shown in the figure is equivalent to sixteen XOR gates, each with four inputs ^ Brother ~ Brother 25B, the use of the x〇r function It is also extremely symmetrical between disk readout and:. Fig. 25B shows the same four-disk operation as shown in Fig. 25A's array, data paths 40, 42, and so on. In this example, the attached # 2 system is not for disk readout, and the drive system to M body # 2 has failed. Therefore, the corresponding data ... 4 (which has no effect) is shown in the dotted line. The second function is different from the remaining data drives (actual materials # 0, # 1, and # 3). 48 200535609 and from the redundant disk Data of physical machine (physical port # 4). This calculation reconstructs the data it stored on the failed drive, and the result is carried through the data path to logic # 2 to replace its currently unavailable data from the failed drive. Μ —The previous paragraphs show some of the real two relationships that may exist between a group of logical ports and a group of physical device ports of a ruler controller—in general, a high-performance RAID controller system Forced to: ST arrays shot by a large number of inventory devices connected to various subgroups of its corpus body 4. One aspect of the present invention is the use of a new type of mapping, that is, storage as associated logic to enable software configuration of the storage device array and improve performance, as further explained later. According to an embodiment of the present invention, the structure is shown on the day of the% chart-the configuration of the shooting register 24 is controlled by 7 of the logical and physical data ports. (The mapping register also provides other feature disks discussed later.) The state example 'mapping register is composed of five stops, both of which are for each of the five logical data. Here is an example. The corresponding block of each logical data port in the register contains the number 1 which is the data itch. The second resource for the logical data slot Γ :: It is pointed out that it is associated with logic "Chun." Chun. The values of the next four stops are identified as ppu,, PP-L3, and PP-M. The fifth logical data port system-a virtual (: r.) Chun.pp The -l4 value is used to specify the physical data port used for the in-position magnetic stele machine. * One of the mapping registers can be of almost any size. For example, a 49 200535609 eight-bit field system An array of up to 256 physical ports will be supported. In the example shown in the figure, there are only five physical ports, and a three-bit block is charged with 7 knives. The five blocks are properly encapsulated into a ten. Six-bit register and it has one bit marked with "r" in the schema for "guarantee. Any type Non-volatile (non-volatile) memory for storing the system can be transported register mapping information.

To demonstrate the use of this mapping register, reference will briefly be made to each of the configurations described so far. In Figure 24A, note that a mapping register 24 is not displayed. The value of Pp-L0 is 0, which indicates that the logical data port continent is the connection port # 0. The following three values are 2, and 3, which indicate that: The two logical data ports of the last two are connected to the next three physical data ports. The value of -4 is 7. This is not a legal entity number for this instance. The value "7" is passed on with 4 daggers. The heart is used to express it. This configuration is not equipped with the same magnet. The specific value selected is the cockroach number. ^ Important 'As long as the material-the actual physical data port 1, 2, 4, gold na \ "" / "system is supported by logical ports 0 to 3. D" 7 "is pointed out--parity disk The machine is not used. Brother 2 4C picture shows the array that is configured for a plug-in-mapping _six_ 'drive. Note: · Logical data port # 2 is associated In the same actual material — it is the same as logical port # 0 于 第一 实 邮 #: a logical port system transfers the data to the monthly port period, and the next two logical port periods. Figure 24D of the Second Reality shows the mapping configured for a single channel scenario: 50 200535609. Logical ports # 0 to # 3 are connected to consecutive ports # 3. All of the 24 pictures are transmitted to the entity The intersection of 3 and its benefits ^ — the different configuration of the data path display.…; Several tribute logic and the 25A map is shown for X〇R € # 〇 as shown in Figure 24A; from all four The data of a logical data port. / R is calculated in the logic of the device # 4 Field example: It is stored in the mapping temporary storage as "7" -... pointing out: i, system "4" instead of connecting to port # 4. Zhuji and it is attached to Figure 25B is a disk drive showing the same material for the job logic diagram of the disk readout direction as the 24th gold material path The system is currently invalid. Logical: It has been connected to the entity capacity M2 system and has been replaced by-"5, Ά; ^ # 2 within the seat to 4. "S" is always one /. The serial port number of the method is 0. Lost W: Remaining value. Its application is to indicate:-the drive is already-lost. Access the output of virtual physical port number 5 and 2, to obtain its data. The baike port system will discuss the self-path switch in the previous discussion. It has been shown that · the four values temporarily stored in the mapping of j, t to a map can be used to represent four logics: to five entities璋 Η, or ~ machine array = month dagger configuration, with or without a redundant disk drive; array of disk drives, with or without-failed magnetics ^ The content of the redundant mapping register is How to use to configure hardware two = sex. In other words, the following discussion proposes the details and configuration of one of the currently preferred implementations of the data path switch 26 51200535609. ',' ,, and σ are explained in Figure 27A of the mapping register. Each of the four logical data is received from five physical data ports. It must be able to connect the failed disk drive from the magnetic stele data. Or if it is a source of capable data, the material of the second entity: the data of LOR. It has six cans-each of them has one (S1X-t〇-〇ne) type multiplexer 5. The multiplexer 50 of 1 to 珲 # 1 is shown in the U-th width. For logic logic ports # 〇, # 2, ## 3) pass the picture of the gate A, but others ( For ^.) Are the same. The selector or "s" of the multiplexer

The car is connected to the logical storage port # 1 of the mapping temporary crying state, “The value of 0 to 4 of pp PP-L 1 is the selected asset, respectively. 4. "" selects the output of the disk read XOR. 27. Figure 27B shows the disk read logic 52. Disk read x〇R logic 52 is a five-input χ R circuit. The preferred embodiment is sixteen bits wide (corresponding to the attached drive data path). (This is equivalent to ten XORs, each of which has five inputs.) Each of the five inputs : For example, the corresponding AND gate (also 16 bits)

Wide) logically qualified (quanfled) or "gated". (This series is equivalent to ten 'NAND gates, each of which has two inputs.) The five N money D gates are qualified by the corresponding five physical port selection signals pp0-sel to pp4-sEL. . The generation of these signals will be described later. The data path to each physical port can be from any of the four logical data ports, or from a disk write x〇R. Examples are shown with reference to Figures 24A to 24D. Although one of the mapping register's slots identifies one of its corresponding data corresponding to the data source of each logical data port of 52 200535609. This asset = for each actual stop. Mapping temporary crying — _ ', σ wait for the side; each of the fields with a “eight,” ^ Dou — Wu — hexadecimal coded field is a kind of eighth — ㈤ decoder The decoders decoded in the logical port block are not shown in Figure 8

Ll-PO, LI-Pl, li P2, Ll PP-Li are decoded into Hp 7 from a source to a destination 'wherein' these names refer to the two paths from logical to physical. For example, L1- P2 refers to Figure 29A. The sampling circuit is from a material port to a physical data port (# (^ 的 "^ 用 = The logic multiplexer 72 of the physical object is markedly industrialized. For four cockroaches) The mechanics are the same. Each multiplex is targeted at the other and close 74 (both are 16-bit should). 72 is the OR gate of the OR gate 76, which has five-and / or arrays and is fixed by adults. (The six AND gates of the An, those who are and the son-in-law are equivalent to the sixteen openings, and those who are long-time ΛΓ turn in. The 0 is equivalent to the evening work , The corresponding output from these logical resources and port # 2 is qualified, that is, the gate system is decoded by five, and one of the arrays shown in L4—P2 is called the machine array〃 # # 料 之 — 彳 目 # Control. In a 4 magnetic array, a given physical port is connected to a disk. Although it is in different cycles. The 66 series of different logical ports has an enable turn. εν ”, brother ^^ picture, each decoding is for the two-drive configuration 'only: D grid has all its outputs. Shi Yuji data port # 〇 and # 1 of 53 200535609 decoder is enabled in a first Cycle, and only decoders for logical data ports # 2 and # 3 are enabled in a second cycle. For this reason,-the secondary system will qualify as the only one between AND in Figure 29A. In other words, , ^ $ The data from the specified logical port (according to the mapping register) is the physical port input =. Focusing on a single disk array, where a single entity does not receive all four logical ports (see f 24D (Figure), only one decoded% system is enabled at a time, so that only one Nazha 74 system will be-inadequate: = Selection-unique data source (logical port). Other open questions are directed to = The source of the "PPn_SEL" signal in Figure 27B. Figure 28 shows the use of a five-input OR gate 68, which will determine that it is directed to-the physical port "" Part 2: SEL signal, if a data path exists in the subject Between the real two. This system provides an indication: the entity bee system functions and can be expressed in Figure 27B reads the disk x〇R. 乂 and 窎 'According to the ΑΤΑ / ΑΤΑΡΙ standard' to send commands to the disk drive system needs to use Program 10 (PIO, Programmed 1) mode, 1 support only PTO f々 Λ /, may be for a device with a pin == 4 such that each access is ~ η ... and is not better than each access ns. The previous command system requires eight or more accesses. If It must be based on the drive system of some AA-8L drives, and the sequence is 〒々. This time is multiplied by the number of drive wait times to the entire process. These commands "port can be issued from A u i" are issued simultaneously by each port-a separate controller, but this is a significant cost. F3 adds complexity and 54 200535609 Tian Bei material in dismantled in-the magnetic part of the array will be located in each drive? 'The given dismantling department for each drive's resources. Eye-to-position. This system makes the address,

Logical Buffer AdH,, and Ier Address (LBA) are the same. The ... The command of the two scholars is to give a given teardown to the 果 ° slave slave. All drives in the car are the same. In addition, it is possible to write a dismantled command slave processor (for example, the same is the case. This is to make the processing of the area and the brother 23 pictures will otherwise send a life of eight ^ Α Hour 20) may not exceed Its broadcast, and common order. "As soon as it was pointed out, a disk array can be composed of a magnetic disk drive." (An advantage of the attached device, or ㈣ "of the present invention is that it has the ability to easily configure a new configuration of these attached drives. The organization j only stores a single array of appropriate configuration bytes. Consisting of an employee's daily deposit port) The enclosing drive of Yiyi * 5 is composed of the attached disk drive. The order (such as item m) can be expected to "spread" to the machine system. First, a subset of the release. These disks are based on "Mask (ma) :: Some institutions must provide — asc) their physical assets that are not related to the current array. The first is to show the implementation of this issue. ^ Bei. DAn refers to Figure 30. The address, strobe, and chip selection signals CS0, CS1, DA1, DA2, DI0WM, and DI0R are displayed for five physical prostitutes (Person and P1). Note: These addresses are all five of the bees of the selected drive. They are individually buffered, making a given magnet: ', "The effect cannot block the transmission of these signals to other drives. Tea Read: Buffer I, 82. For—The given chip selection information 55 200535609 No. CS # 0, CS # 1's drive-out driving force is targeted at the far 86. Not mapped temporarily Register's current

The signal station is qualified. Refer to any of the spleen spleen selected by the content of Pn-SEL of Θ, 84, Θ, ^, or any one of its chip selections, and therefore the read and write gates will be ignored . It seems that it is meaningless for the general item to be returned. The conflicting data value is returned to-total /, for the heart. / For example, a "general readout" is caused by the implementation of the second party and the item.

Pn—DI〇R #) is "Propagation, to the place 2. Pui out of control (Yu 30 choices of pictures (Pn CS0 #, pn CS1 # ^ People as early as the daily film selection-##) The attached storage devices will return the data to the physical port, which will be the falling edge of the material. And I will create a circle; operation: 1 lock on Pn-DI0R # gate to cry ^ 1 P to return —The values are processed to the region as a result of this read cycle. A region processor will then read each port for the next time, using: the same address 'which does not cause Pn-D1' strobe cycle Repeat and ... ', and must change any lock data. These cycles allow the region to process: to capture the potentially unique values of its data locks stored in each data lock. Pn up to 600 ns may be required— The DI0R # cycle is only executed one time. The value of the lock can be captured for 15 nS, which is better than repeating the Pn_DI〇R # cycle for five times for significant savings. , General read "and" general write "f sets allow the area processor to send commands to the currently selected array and receive them in the smallest possible amount of time: from The control status of the Hai array. When a different sub-array is selected by loading a new value into the mapping register, the 'control interface is automatically updated without other code changes. 200535609 Sad Order-Order (Status Ordering) There are many entities generated by the system processing, and it is not how they can be promoted by the mapping register. Each test of these ports also has multiple input signals. Again, these signals are associated with the logic bucket H system It can make the software ^ € (overhead) ^ t ^, ^ ^ # „t cut off the output. Its use is to send a message to the service diagram from the controller to show the use of a multiplexer 90. The multiplexer crying ° y ◦ i By mapping

The interruption controlled by the PP-LG value of the register to select its physical entity associated with the logical data. Each of the other logical data ports has an identical multiplexer (not shown), which uses the corresponding PPjLn value to find its interrupt. In FIG. 31, the buffer 92 obtains the selected interrupt from each logical data port multiplexer (90, etc.). When the area processor (20 in Fig. 23) reads the interrupt status through this buffer, the interrupts appear in the order of logical data ports and logical data port 0 is at bit 0. The same technology can be used to sequence the internal and external signals from the physical data port, including the drive cable ID signal and the internal FIF0 status signal. This feature results in regional firmware being able to use a common sequence of codes to target multiple arrays with different numbers of physical ports. Once the interrupt buffer 92 is loaded, the required status bits are always "sorted" for any selected array, the least significant bit of the register. The number of bits can be hidden and reduced As small as the actual number of ports.

Interrupt ANY and ALL As shown in Figure 31, the interrupt selected from these logical data ports can be a logical AND (at AND gate 94) or OR (between 0R 96) 200535609 to provide the signal "Interrupt Kan ,, and "Interruption,". When the area processor has issued a command, and before any data is transmitted, it may be necessary to interrupt the request from any (Aγ) drive, because one or more of the disks may have been rejected The command may have some other error. To find the drive to start transmitting data, the regional processor will want to know when the drive has determined its interrupt signal. Because of this: ρ 7 is all right. Note ... This type of implementation is such that the software is on the number of drives. (For an array of two drives, the interruption signals from each device were recognized by the elder brother, but the array of the earlier one.) The same drive works. A The AND and all signals are still positive though: there is a need to access individual devices for initializing Erlifang on the use of the soft system for part of the execution time; special errors in the device. The needle appears in the address of the area processor, inside the door, and the access of each physical data port to any one of them: solution; special position. For these temporary caches, the Ai and Ma horses, the decoded output is re-mapped according to the mapping: <Uchitani, loaded with a ', ..., during initialization' mapping register. Tied to physical port 0, logical clothes! = L means device 0 is pointing so that the physical ports appear sequentially, from σ to shell body port 1, and so on. This system ^ ^ 4e starts at the user's address space in the processor's address space. Yu Zhenggong's younger brother, a physical drive map. If the register is to be loaded-logical to, the regional processor is "taken ::: =" is a self-logic ... disconnected disk drive, through the unique address 58 200535609 T, when the identification picture is Loading and accessing real and strict. One—Ke Ba% called, Ding% [馐 ping ζ. This order service is not related to the characteristics of the physical data port address to which it is attached. A hardware implementation is shown in Figure 32. When the processor accesses the address area of the device's cockroach space, one of the eighth type (r, e-of_eight) decoder '100 is the processor address. Lines five to seven, the meaning of which is directed to the space of 22 bytes each. The corner of each space, Wu Shima, determines that the system _ 庵 七 的 M £, soap N decoding signal Pn-DEC. The decoding of the virtual port number seven is aimed at the signal of -1 shovel, access, and access. The P7-DEc signal and / or the port selection of each of them to decode the alum number m / Yunli 102 , So that the resulting drought Pn_SEL (㈣, is determined for the material and for-universal access. Special access

… 璋 Each of the selected signals is then pushed by the "j" value from the mapping register. The one-of-eight type decoder 10 signal and the PP u # P2 from the mapping register H are used to generate its The form is L2-p0_, and the signal is from the logical port-Zhibaibai, and the alum number, which points out: Come, bait-comes from Jingtian wood, one of the zeros in the body. These eight-type decoders of the four logical ports are the same: for each of the other physical port systems, there is a five-rounder (not shown). Used for the display of the 06 of the physical objects. For example, ^ ... [【 -Five different source systems for film selection :: one of the body # 2 pairs early-sub array of the disk drive, chip selection ... meaning ... by the knowing device, but for all of the dual disk drive The four logics are determined by both of these logic devices. ^ The chip selection is described in the foregoing and in the drawings, the number of examples of the model 59 2 0 535 609 is illustrated; it is for each logic magnet: : Of — a kind of logical mapping register. As explained, the column 'is in the abstract position, the -value bit is provided —The defined matrix is called an entity mapping with a corresponding physical port number. For each physical port or example, a register is provided with a corresponding logical port number. H 'bit, -value It is related to the following example. The generation of the mapping register is to explain the disassembly-finding sequence and store it in each of the brakes. The process is repeated for the helmet and the disk drive. For example, the Flute Factory block of data (and the fifth block, Wang Shi F〃, the younger brother of the area 〃 £ ghost, etc.) are stored on the drive connected to the physical block. The second block (And the sixth area only, etc.) are stored in 1 connection &amp; ghost, ㈣ of the drive. The third block of data (and the seventh block, the first n ~ to the real real distance # 4々 ten blocks) , Xunsi) are stored in its connection machine, / the second "disk drive. The first block of data is performed on the logical disk. The block is on the logical disk 胄 1, and the third block is on the logical disk. 2 and the fourth block is in logical drive 3. The alternative type of mapping register for this example is as follows: logical mapping

Logical port # 2 1 value (physical port 4 2 60 200535609 Physical mapping physical port #-~~ -——----- 4 3 2 value (logical port — 1-— — 2 — 1 iv. For magnetic Narrow and wide dismantling of the disk drive controller, partial disassembly updates and further improvement of other synchronous operations

A · Introduction -u D Μ A With the continuous development of the ATA / ATAPI interface described earlier, the PI0 data transmission mode is replaced by UltraDMA or _Α transmission mode. Mode performance is limited by the interface and winding back and forth ㈣ カ ㈣ was delayed. The UDMA mode uses the same electrical interface in-"source synchronization," mode, where the connection end of the source data (port or disk drive) provides the ^ = strobe: by the source synchronization clock, the performance is mainly affected by It is limited to the skew caused by the early conversion of the electric continuum and is not limited by any round-trip related delays. Although this enhancement is to advance the data transfer rate, the MBPS can be as high as 100 MBPS or soil. Synchronous lean material transmission and cutting system applied by the traditional ATP will no longer be applied to the timing of the origin of the magnetic disk reader system for this direction. The reason, above the disk, use the FIFO memory to modify ΑΤΑ / ΑΤΑρί The UDMA version of the interface is compatible with synchronous data gates. The 11th Sihan Fang and the device system have been described. The advantages of synchronous lean data transmission are also discussed above. B. SATA FIFO- With the ATA / ATAPI interface Establishing one of ΑΑΑ or SATA for a parallel interface is replaced by a point-to-point (P0ini-i0-P〇im) serial interface connection. In the serial interface, there is L · System and Data Transmission System 61 200535609 Sash information structure (iv), Frame Informatlon Structure). Each ^ S system has a 1-header, which indicates the FIS type, Payload, and one of the CRC fields μ ,,,, and Tian used to check the integrity of the received FIS. The system is defined in the interface specification. Fig. 33 is a simplified overview of each embodiment of a disk array controller according to the present invention, κ ^ ~ ° uses a plurality of SATA ports and disk drives. Yu Di 3 3 f ^ l, 4 η &gt; a., ,,, is a concept display of the 420 series of impulse states. In fact, Bu Xiao's buffer system can be used by Keli Tianzhang + / This 4 J uses any address space of the host system's address space. The illustration shows a single point. This, the friendly state 420, the host interface 450, the direct v, the physical access (DMA) channel 42 times. 丨 1 access to the day, 422, the shell material path switch logic 46, and each disk drive 428 / 、 As for the miscellaneous one, 〃, car 424. However, the current "relationship between each channel 460, data mode, M drive 460 depends on the data path switch logic projection plan ^ As detailed earlier, almost any data map c eme (for example: Redundancy, disassembly, etc.) and any reference from real disk drive to logical disk drive: physical disk path switching logic 460 is implemented: Hunter reconfigures data path reconfiguration (for example, ... (One when j is dynamic ^, when the disk drive fails), and it can be an embodiment that is also a good support when the soft f path is opened, and the data path of the data row, that is, one day machine and for multiple arrays Involved. — When the temple path is set, no software is needed. Figure 34A is a more detailed host interface 450 of Mingdi 33 here. In this example, the host interface is included in the ritual recognition system. Bridge 500, multiplexed for the next 4 pass but fortunately different from the system main memory and a ', (, and

Ci Xinjiang ranked between 502. PCJ 62 200535609 The row system is just one example. The PCI bus is coupled to various logical direct memory access (dma) channels 510 via the -ci bus interface 504, under the control of a bus arbiter 506. In operation,-the second time is that only the DMA channel is actually transmitting data to or from the ρα bus. That is, multiple DMA channels can function as the same temple, as further discussed by Lai later. Figure 34B shows a SATA 璋 52 in more detail, indicating the operation in the read direction of the magnetic disk. According to the published SATA specification, the interface is implemented:-physical layer 522 for physical connection to-attached disk drive 524;-link layer 526; and transmission ^ 528. The transmission layer includes a first-in, first-out, 0) memory胄 530, which provides a data buffer between the physical disk drive and the controller. The SATA interface specification provides a handshake mechanism for connecting the ends of the 532 to suppress (suspend) the other end One of the data transmission. SATA connection is half-duplex. This protocol uses a reverse channel to exchange signals for each FIS transmission. Due to the much higher speed of connection 532, up to 80 additional byte systems It can be received when the request is suspended during one of the transmissions. When data from the drive is received, the FIFO of the transport layer 530 can generate an "almost full (xian delete M)" instruction (not yet Show no) '纟 will use the reverse channel link to Prevent fif〇 overflow (_ 心 w) and data loss. The data path 54 on the other side of this FIF〇 can be accessed for the timing generated by the area. In the readout situation, the control flag "EMPTY" from ", And the control signal" p0p "to FIF〇 is used to control access to data. 63 200535609 Figure 3 4C shows the same wATa system signal" FULL, "which is displayed during an official, disk drive insertion. , And "push," is used to write the shell material to the Haihai FIFO 'for subsequent transmission to the attached disk drive. Furthermore, the data transmission system is decoupled from the actual link to the disk, The transmission wheel system from the controller can be synchronized, and the lever effect should be "real-time ,: the advantages of the remaining operations, the pot is the vertical weight. ^, A few 460,", "the switch logic of a preferred embodiment System structure ==, the 33th embodiment 'if the switch logic is biased to use two data drives and a parity control device / Han Jie Xi Xide, the master ja π into a control ^ The transport layer FIFO is accessible for a common strobe for calculation by the ship. In this way, the original is superior V υυ Passing wheel c. Data path switch logic is assigned to the data path switch logic path indicated by the dashed line 46G in Figure 33 ^ The SATA drive port and the DM that accesses the buffer memory are connected to 3.2 The data path switch logic of one embodiment is described in detail above with reference to the Mth section such as user Λ. For example, see: Data Path Switch%. It is provided in the logical data port or DMA channel and the physical data is mapped; Γ state Configurable data path. In one embodiment, the configuration is determined by the data that can be stored in the mapping register. As mentioned above, the mapping can be logical or physical. The = 40 diagram is a simplified diagram of a data path switch. Configuration, and display four DMA channels _Α0 · Α3), using, host or buffer memory data transfer wheel. The open relationship preferably includes 64 200535609 hardware for implementing synchronous transmission and immediate redundancy (XOR) operations. This block is inserted between the DMA channel and the SATA port (SATA PO-SATA P3). The specific numbers of DMA channels and SATA ports are for illustration purposes only and are not limiting. Although some other diagrams show specific connections between DMA channels, SATA ports, and XOR logic (for example, Figures 13, 14, 24, and 25), these paths are logically configurable, such as Discussed earlier. All data paths are preferably 32-bit for SATA and 16-bit for PATA / UDMA applications. • Refer to Figure 40. Disk writes to any SATA port can come from any DMA source A-D or from the XOR box 4010 output "X". As indicated in the figure, each SATA port has a FULL status flag output. In order to transmit data synchronously, the PUSH signals of all active status ports are determined at the same time, but only if the FULL flags of these active ports are incorrect. XOR block 4010 can calculate the XOR of any combination of DMA inputs. By qualifying the appropriate combination of these AND gates and enabling logic signals (eg, XB_ENA) as necessary, it is targeted at a specific disk array. ® Configuration or disassembly plan. The DMA channel data path may be 32, 64, or 128 bits, depending on the width of the data path to the memory. The data path switch block will package or unpack the 32-bit SATA component as needed to establish the data path width. For the narrow disassembly disk writing situation discussed below, only one DMA channel is used. The logic receives DMA-wide data blocks, and it outputs 1 to 4 32-bit blocks at a time, which is synchronized to the disk array and XOR logic. For the case of a narrow disassembled disk read 65 (S: 200535609), this block is used to synchronize the square i-elements of the rich and fine Sayada 1 and receive 1 to 4 32-bit sub-groups at a time, using as discussed above For wide disassembly applications, there may be more than Y SATA # deleted. For the purpose of the "host interface details" in the figure, the factory is used. As the wth operation is the same, but the main difficulty is ^ The master ^ LL and the cord are transmitting one data at a time. In this form, the lean path switching logic is directed to one of the DMA channels. The block (Figure 40) must provide memory 4020, allowing it to be-, Buffer or first-in-first-out (delete) At the same time take out the magnetic 磾 i # 彳 /, μperson ", and enter one, but also allow data to be taken out on the drive side. For ^ / FIF0 series must be full size to allow ^ System performance, these buffers or multiple complete caches rca H ^ Allow the host bus to transmit data for all cache lines at once. D. Disk Write Accumulator Figure 42 Explain that one is used for _ road. The x〇R accumulator electric t_ ^ Λ '° Λ, the integrator system—a substitute to provide additional buffering for each DMA that is required for the purpose of the pass data. Shared host bus mediation: the host pass from different buffers The first advanced mountain n will never be regained. In this legend, the FIFO s memory is 421. For example, this F t hurts π &amp; ', and the length is a magnetic zone. Λ μ ^ Recognize and write a complete dismantling of three assets "\ plus a few more drives 1 h J 丨 early evening J, use the following process:! · This process will wait until NOT FULT # l Some disk drives are used to indicate that LL 'its in this case will be ^ ("^ σ · It can accept another magnetic zone 枓. ^ Read: take the L flag in Figure 40.) 2 Next, a magnetic field shell system is applied along the line shown in Figure 40.

CS 66 200535609 DMA channel 0 of path "A" is transferred from buffer 0 to SATA port 0. At the same time, referring to Figure 42, this data ("Α") will also be enabled by Α_ΕΝΑ through XOR 4222 to FIFO memory 4210. 3. Next, the data of one magnetic sector is transferred from the buffer 1 to the SATA port 1 using the DMA channel 1 along the path "B" shown in FIG. 40. At the same time, referring to Figure 42 again, the data "B" will also pass XOR 4222 for enabling by B_ENA. First-in-first-out (FIFO) memory 4210 is currently enabled by X-ENA. Therefore, "B" and "X" are XORed and the result is sent to first-in-first-out (FIFO) memory 4210. 4. Next, the data of a magnetic sector is transferred from buffer 2 to SATA port 2 using DMA channel 2 along the path "C". At the same time, this information will also be passed through XOR for enabling by C-ENA. The current contents of the FIFO will be enabled by X-ENA. Next, "C" and "X" are XORed and the XOR results from these magnetic zones from buffer 0, buffer 1, and buffer 2 are transmitted to SATA 3 synchronously. Different from adding FIFO of one or two cache lines to each DMA path, the above plan uses only one FIF 〇 and the length is a single sector. The process of moving a magnetic field from each buffer will be repeated as needed until all data is transmitted. For example, a host bus rate of 450 MBPS will support a full 150 MBPS for four SATA drives. The above is directed to a complete disassembly write. This series can be purely a parity drive update, if it is a transfer from A, B, and C to SATA drives 67 200535609 For exemption, only parity transfers are reserved. E. Data path switch readout direction The data path readout direction is illustrated in Figure 41. Again, this block is inserted between the DMA channel and the SATA port. Although other diagrams show specific connections between DMA channels, SATA ports, and XOR logic, these paths use logic similar to the above and are actually configurable. All data paths are 32-bit for SATA or 16-bit for PATA / UDMA. _ Disk read data from any SATA port (SATA PO-SATA P3) can be advanced to any one of DMA destinations A-D, as indicated by the concept of a multiplexer (for example: mux 4 1 1 0). In addition, the data from any SATA port can be input to XOR 4120 and selected by the corresponding enable signal (for example: X1_ENA). Each SATA port has an EMPTY status flag output. In order to transmit data synchronously, the POP signals of all active status ports are determined at the same time, but only if the EMPTY flags of these active ports are incorrect. X0R Block 4130 can calculate the X0R of any combination of SATA inputs, and qualify the appropriate combination of these AND gates by β. The DMA channel data path may be 32, 64, or 128 bits, depending on the width of the data path to the memory. This block receives components from these SATA ports and can receive components from X0R to replace data from a failed drive. Ingredients from 1 to 4 sources are packetized to create blocks for DMA transfers. For wide disassembly applications, there may be more than one DMA channel. As shown in the "Host Interface Details" diagram in Figure 34A, these DMA channels can operate simultaneously, but the 68 200535609 host sink muscle system will only transmit one data at a time. In this case, this block must provide a small buffer for each ship channel or allow the transmission of data to the host one channel at a time, and allow simultaneous transmission of data to the SATA side. The first explanation is 414. For good system performance, these buffers / F systems must be sufficient. + ^ ^ Allows the host bus to transmit in one transmission or multiple cache lines. E. Disk Readout Accumulator

# 图 系 „Erming is an embodiment of a product that accumulates a pair of data path switches in the readout direction :: road. The accumulator is an alternative phase channel with two identical = required extras of each ΜΑ. Buffer, because these are delivered with A 34AFI ,, and the materials are shared. The shared host bus interface (see Figure 3 4 A) is to ensure ·· 炎 白 # ^ ^ H, 5, and 11 money machine transmission system The first-in-first-out (_) memory side that is not duplicated in this illustration is a magnetic area. 』％ 食 戾 读 二 置 4Γ is used in many ways. For example, it can be used in columns and under 徇 ... Disk drive plus redundant disk drive array SATA2 disk drive is invalid. Use the following process, κ This process will wait until all the-. Shu earn TY, which in this case means: It can provide = to point out that 2.- magnetic field data is transmitted through DMa through a magnetic field. To 'buffer 1.', refer to Figure 43. Simultaneous transmission of the "Enable from SATA 0 / brother 43 picture" 'From first to first out (FIF0) memory Gane. … To pass the job side 69 200535609 3. Then, the data of a magnetic sector is transmitted from SATA 1 to —buffer 1 through DMA channel]. At the same time, the data from SAATA II enabled by P 1-ship will be in the same position as the advanced first memory fV enabled by X_ENA, and the result will be returned to the FIFO. 4. Next, a magnetic field data transmission was enabled by P3-EnNA, ΑΑΑ 3, and again the current content of FIFO dfe, 脰 4310 and month b caused by χ —

. , Extend XOR. The result is extended to buffer 2 through DMA K to complete the complete disassembly interpretation. For a complete disassembly of a failed drive, it is interpreted that all data that was not lost is transmitted first and then the same drive data. For the missing; two == one block read, the corresponding area results from three ... ports will be transferred to a buffer for -DM "f lose XQR'_ with the final ^ DMA channel. Unlike adding one Or-this M a path, the FIF of the upper, ^ + ± | lines are connected to each DMA road and there is only a single chat area. Since each SATA is described, the length is-one magnetic .. &gt; Λ. Drought movement—The process of a magnetic field will repeat as the snow demands, until all the data are%

The main feature is the example of "Stop soil and lose cattle, and 450 MBPS. The 4 rate system will support 150 MBPS for three additions." Το G • Synchronous readout method of the indole-general case_ Just by way of example, Zidou, L Beijou is disassembled into 4K wide than three data. Plus a redundant disk drive Beckham disk drive is set to -4K read access, basin 4 4 special cases are false. ^ To be consistent with a specific disk drive 70 200535609 A specific disassembly 4K P said, Naoda π Dragon boundary ' However, in general, the number of data inscriptions will be as long as there is a redundant drive. The width of the solution for any glacial soil for a single magnetic field is L-numbered. In addition, a read request requests any magnetic field, and a guide sound 仫 μ, ′, can start from /, and can target the volume of a column from a single magnetic field. A general method for synchronizing the data in the array according to the present invention is as follows: 'Machine array = block is a disk drive containing the initial sector, ... The end of the block, or the read request = from: Hai on whichever comes first. # # 、 、、 口 束 None The wood block is right. The beginning of the magnetic field happens to be in the lifetime, and the corresponding range of magnetic fields is read.

And "immediately, transfer the results, such as the disk drive, calculate the additional entire block of data requested by x〇R _, read out the continuous blocks from y and the continuous disassembly. Transfer Performance data, the access system must not occur on each drive. When-the block is a drive that requires self-loss, while the non-failure magnetic system at the same time needs to be "u"; since all other drives The disc player's data is known to re-create the failed block. Final'-data read access is made by including the drive where the final position lies. In particular, the U and B materials are transmitted from the beginning of the block or from the initial magnetic field (if it is in phase '^ to the final magnetic field. If the block of this block is located in the position M.) ^ ^ Λ, r ^ Disk drive, then read the corresponding magnetic zone range of the disassembled block of other disks, calculate it by the ship system, and send the result back. The former &amp; It is enabled by various devices, especially the data path switch disclosed in this article. 71 200535609 The machine update system will also need _ redundant disk drives. More :: What data disks have always properly reflected such data disks. The contents of the disc drive. The redundancy is updated to update the array. For the array: for a complete disassembly to receive new data, a standard RAm x〇r system data block is a data drive. Buffer and capture the written data into the engine can be used to generate redundant disk drives from various components of the new internal two :: binary # X〇R this to-buffer for a subsequent write transmission to Redundancy 2: Write = Example of-4K wide disassembly, except for transmitting postal materials-the same 12k Material m + έ, section to disk drive, "from these buffers and uses redundancy of 4K for additional data transmission totaling 16K.-Update can involve less than-complete disassembly. However At the time, the contents of the blocks of the redundant disk drive were all unresolved, and all the new blocks that were dismantled were replaced with the previous blocks that have not been updated. Figures 37A to 37C are for Partial disassembly and update-a method of prior art 'is also called-a kind of read / modify / write (R / M / w, Read / M. Ring / w_) operation. Figure 37A shows the display steps}, Read unchanged blocks from a data drive and read the corresponding blocks from a parity drive. The data and redundant data are stored in the buffer 3702 respectively. , 37〇. In step 2 of Figure 37B, by reading out (and performing duties), the new warship contains a buffer of old data, old parity, and new data '3704, 3704,

Formed with 3706. X0R results are stored in buffer 3708. Each buffer has a corresponding DMA channel as shown. Finally, at 37C 72 200535609, the new data and the new parity data are stored in the separate 3 disk drive of the magnetic map. Another conventional approach to this problem is to read out in advance that it is an unupdated block. At this point, all new blocks and unchanged blocks will be available in the buffer and can be read out for X0R calculation. Read-ahead is to establish the same initial state as-completely disassemble the writer. In this way, ^ data drives are read or written, and redundant drives are written. Figures 38 and 39 illustrate an embodiment implementing a modification of the partial disassembly update. In item 38 ′, the first is to read the current redundancy, which is the ^ OR connection of Mm_ of the current content of the block to be updated ~. (X〇R is implemented in the array switch logic 460 0 as discussed) The result of this calculation is equivalent to the XOR of its ^ block. This x〇R of the two blocks counts: all of which will be the effect of the current data of the updated block. ~ The intermediate result is the 39 figure of Jiancun McCord. This intermediate result is then XORed to generate a new redundant block and a new data block for the entire dismantling of Beco 3812: By. Update 3 820 and data drive 3822. herein:. To several (parity) disks Two XORs and two writes. "With two read, L redundant write-synchronous data transfer wheel _ completely disassemble the write operation system can also operate-a kind of synchronous redundant data transfer wheel and two" for one disk read * magnetic The disk write command is issued to each drive in the array as follows. 73 200535609 When these drives are ready to transmit data, Dw q (engme) grabs a first component, X μR of 1 μ ^ from each buffer, and then writes these first — Write the results of the calculations to the redundant disk drives, using the DI0w that is shared for one of the parallel ΛΤΑ / ΑΤΑΡΙ interfaces. "" As mentioned above,-the introduction of FIF0 into each drive allows the same method to use 'Although it is DIOW for a pusH, for its use of the ATA protocol or SATA magnetic Disc player. ② The advantages of this method of disc player include the following: · Transmission—The total amount of data from the buffer is a new lean material written to the magnet. The extra 2K buffer is written. . Youchong ... Bei Chu and 4K's buffer The production of more than a few data has not increased DMA engine injury Congdou's temple. The use of special… 糸 will involve the redundancy of writing calculations again, making it noisy u u ~ 巧巧 、、 友 衡 杰 U and。 Be ready for transfer to redundant drives. As discussed above, a part of the disassembly and update system is updated. According to Boguzhi, it is disassembled as one yuan. It is read in advance that it is unchanged. The inner valley of the block that is only right ^ | + 读 出 is read in advance. By the same token, there are several tribute materials for Yin 1 Xia who want to compete and use synchronous redundant resources ... The described method of the block is the advantage of transferring more than a few rounds, as follows: Buying tfp makes the first release machine. More than a few disk drives and magnetic disks to be updated2. When two disk drives are ready to transfer wheel data, m from each disk drive Calculated and the results are stored in 74 200535609 in a buffer. This process is repeated one by one until the complete block is read from the drives, and the resulting block has been stored in a buffer. This buffer holds XOR of all blocks that it has not updated. 3. —Write command is issued to the redundant drive and the data drive to be updated. The disk drives are all ready to accept the data. 4 · Dangbei / 丨, Ding is captured from XOR 缮 | 突, η is also ^ /% said AUK, You Chongwei and a first data is captured from Update the buffer. Two ingredients eve vnp ^ The XOR of our knife is a nose. The data is then simultaneously transferred to the two drives for the use of a shared mow strobe. The data drive receives the components from the unchanged update buffer, and the redundant drive receives the calculated XOR of the two components. This process is repeated in steps S 'to complete the partial disassembly update. Note: In the last step, the two drives need not be written synchronously. 4 XOR and the contents of the redundant buffer ^ Γ To perform XOR and write 5 π excess disk drive without self-refresh buffer "', This is the transfer of w 亀, but J • synchronous read method- Narrow teardown_ For this point, the synchronous data transfer wheel &amp; becomes "based on the teardown width of one of at least one magnetic zone. This is a data structure supported by it. These dedicated _ engines ^ ^ ^ The DMA quotes the total μ data and operates as it has been transferred to the buffer. Above, [1 is based on the fact that once the reader and the method are equally suitable for its less than—the synchronous data transfer wheel described above refers to Figure 5. , Its description has the data: = the disassembly width. The machine plus one redundant disk drive is one of the two data disks. As before, a π # logical sequence is 75 200535609 Degree system selection. Slanted atA / atapi drive, sixteen of the data path ^ Fang Bu Meng's disassembly width. Magnetic area. The-character group is stored in; Degree system-yes (in Figure 5) Information 〇), 坌 —A. Λ Yu, Di—disk drive 卄 0) 'The first group of xi is stored in material υ, and the third Group (2) is the storage brother—disk drive (the process is then repeated to store the nth drive (data 2). At this second position, the fifth data word is stored. I give, and (4) in the second drive and stores the sixth data block (5) in the second-standing position of the third drive is repeated on a word-by-word basis and disassembled one by one to the magnetic The end of the disc / this process is said to be stored in the parity drive. For example, the :: Parity parity drive word group is displayed ... 8, the second disassembly 2 'drive word group 6, 7, and The position of 8. The data stored in the data magnetic configuration shown in Figure 5 'a complete dismantling is by the data drive from each data. Choice, and Cheng' access from one of the three drives , Even if it is not: useful = the need will come to an end. The total bandwidth of the data drive with the data of the bull ^ is the combination of the word width of the first attack and the step access. Degree, block, maintain the disassembly order, and store in the buffer cry. Body * Kaodi 38 picture, the composition of the word group width in the narrow disassembly machine described above is synchronously taken from the remaining "4 second-effect" magnetic And from the disk drive. The three words captured are counted, and the XOR from the head, the, and the real-time solution is based on the combination of the data of the two = disk drive. The original data is rebuilt and rebuilt. The result is transmitted to the buffer: 76 of the 2005 disk drive. 200535609 The luxury of the configuration here is one of the results of disassembly: One of the magnetic disks # I of the disk drive. The magnetic field request is not for the three, '4 right one (modulo) three magnetic field boundaries, =-multiples and aligned to a modulo block system read, one two: the system will require · three Sectors-Only one or both of them are passed to the drive, and of these sectors ..., back to the host. The magnetic field is ... The complete magnetic field is for transmission, and the required area is discarded. One or two magnets on the mouth of the moon ruler. Those familiar with this technology will understand the main principles, and you can carry out the above two extravagances. Without departing from the present invention, the above-mentioned implementation details are changed. The scope of the present invention is only defined by Shen Zhuan, and thus the profit-seeking target circle is defined. [Schematic description] Brother 1 Α to 1D 图 倍 # 日 η 含 4 ^: u You have various drive configurations. f 2 is the dismantling data of two data drives. Brother 3 illustrates the disassembly of two materials. Fig. 4 of w «plus a redundant disk drive illustrates three data magnetic disassembly. The fifth picture of the ticket plus a redundant disk drive is a narrow disassembly, which is ―width of a block (which is shown in FIG. 6 as a simplified audio chart of a disk array system). UDMA drive data readout data path: Figure 7 shows a simplified schematic diagram of a disk array system in the write data path written to the UDMA drive. Figure 8 shows a disk array socket "Simplified schematic diagram of the path to the shell" 77 200535609 It has "real-time" redundant data storage. Figure 9-Simplification of the read-out path of a disk array shows that it has data "real-time, regeneration, one of which Disk drive failure: Figure 10 is a timing diagram illustrating its disk array read failure. Figure 11 is a functional diagram of a disk array controller that displays the logical data path and DMA channel for each entity. Figure 12 shows the actual implementation. There is a consistent DMA channel and data path for all physical ports, which are provided by 1 series and 1 Letian queue switch. The DMA content of each physical port is stored. In a ram. Figure 1 3 is a description In a green stone &amp; mountain. The data path setting of the array switch that the disk is out of. The figure 14 illustrates the data path setting of the array switch used for a Weiwei,, ^ disk writing. Figure 15A It shows the dismantling of the data of two, two, and seven drives per person. The chart shows the data mapping of one or more magnetic males with one jBC) Ekraid〇 point pair; I: #. 1 mouthful of money, Figure 15B illustrates the mapping of data for RAi T 耵 RA1D1 or mirroring. FIG. 16A illustrates the RAIDXL data mapping for a non-redundant array. -Personal Four Magnets Figure 16B illustrates two or three data disk drives. The last machine plus a co-located disk drive mapping. Some of the RAIDXL data of the disk drive are shown in Figure 17 for-^ T to a belleville disk drive plus a co-located virtual machine or three data drives * A possible 78 200535609 RAID5 data mapping. Brothers 18 to 22 illustrate various array switch configurations and falls. Figure 23 is a simplified illustration of a disk array controller. It is used to interact with a host bus. It is a wide disk drive interface and is used to interact with multiple hosts.

Figure 24A is a conceptual diagram. The logic data 4 of Mu Guzhi, the gate of material port A, is directly connected to the physical assets, and it displays the corresponding mapping register contents. Brother 24B is a conceptual map, which says that the physical information specified by the cockroach contains five organ contents per system and one earphone. ′ And it shows the corresponding mapping temporary storage. The 2 C drawing is used as $ 日 g ^ Each disk drive is a two-five drive array conceptual diagram, in which the corresponding mapping register contents are not. -Page 1 2 4 D is a conceptual diagram of a single disk drive system. Among them, Er 0-3 is a serial register that transfers data to the corresponding register contents. Only the body 3, and the display of "25A" shows the direction of the drive configuration of the "a": the job logic; and it displays the corresponding mapping register contents: write path =:! :: The XOR logic with the same data as in Figure 25A and the output direction of 5 tickets, except that the drive system attached to the physical port is currently invalid; and the mapping temporary memory is displayed. "# 2 As an example, the data between the array controller and the controller M = the register is controlled by the logical and physical resources 79 200535609 Figure 27A is a conceptual diagram of the multiplexer circuit in the read data path . Figure 27B illustrates disk read XOR logic in one embodiment of an array controller. Fig. 28 illustrates the decoder logic of the logical register # 1 (PP-L1) block of the mapping register in one embodiment of the array controller. FIG. 29A illustrates the logic path to the physical port data path logic in one embodiment of the array controller (illustrated only for the physical port). FIG. 29B illustrates the array controller in one embodiment. The disk is written with XOR logic. Figure 30 illustrates the disk address, strobe and chip selection logic to enable universal access commands to a currently selected array. Figure 31 illustrates the logic used to associate with the logical disk. Figure 32 of the interrupt signal logic of the disc player illustrates the hardware implementation of logical addressing. Figure 33 is a simplified conceptual diagram of an embodiment of a disk array system according to one of the present invention, which uses a plurality of sata cock and disk drive. Figure 34A shows the details of the host interface in Figure 33. Figure 3 4B shows the SATA port interface in the direction of the magnetic readout of the ancient σ. Figure 3 4C It is shown on the SATA port interface where the magnetic disk is written in the direction of Gubu Zhuma. Figures 35A and 35B illustrate the read operation of the prior art in a coupon 1. One of its disk drives has failed. Section 3 6 The drawing illustrates a &amp; from the mountain according to one of the present invention ^ ^ A new read operation, one of the disk drives of which has failed. 8 80 200535609 Figures 37A to 37C illustrate a prior art read-modify-write in a serial interface drive array ( r / M / w) operation. Fig. 38 illustrates a new buy-modify-write operation in a serial interface disk drive array according to the present invention, step 1. Fig. 39 ® illustrates the operation according to the present invention. One of the inventions: a new read-modify-write operation in a serial interface disk drive, step 2.

Fig. 40 is a diagram illustrating an embodiment of a disk array switch arranged in a disk write direction according to the present invention. ° This figure illustrates an embodiment of a disk array switch configured for disk reading according to the present invention. ° Figure 42 illustrates an embodiment of a magnetic disk according to the present invention. W⑽ 峰 # -Each system in the ear [Description of the main component symbols] in the magneto-logic of the present invention 10 ·· Disk drive array 12 '20: Disk drive 14: Disk drive 16' '24: UDMA Interface 18: buffer memory 20: control processor 22: disk drive interface 24: mapping register 26: FIFO memory 81 200535609 30, 32 :, almost full "signal 36: logic gate 4 0: Logic block 42: "Everything has data" Signal 44: Read strobe 46, 48: Data output path 50: Multiplexer 52: Buffer

54, 70: counter 66: decoder 68: OR gate 70: data path 72: multiplexer 74: AND gate 76: OR gate 80, 82: buffer 84, 86: gate 90: multiplexer 92: buffer 94: AND gate 96 OR gate 100, 104: decoder 102, 106: OR gate 300: data drive 82 200535609 320, 380: UDMA interface 322: invalid data drive 340, 370: FIFO memory 342, 344: data path 350: buffer 360: XOR logic 390: redundant or parity drive 392: path 394: XOR logic 396: XOR output 420: buffer 422: direct memory access channel 424: SATA port 428: Disk drive 450: Host interface 460: Data path switch logic 500: System bridge 502: PCI bus 504: PCI bus interface 506: Bus arbiter 5 1 0: Logical direct memory access channel 520 : SATA port 522: physical layer 524: disk drive 83 200535609 526: link layer 528: transmission layer 5 3 0: FIFO memory 5 3 2: link 540: data path 620: XOR logic 622: buffer 3702 3704, 3706, 3708, 3810, 3812: Buffer 3820: Redundant (Parity) Disk 3822: Information drives 4010,4120,4130,4222,4320: XOR 4020,4210,4310,4140: FIFO memory 4110 of the multiplexer 84 ··

Claims (1)

200535609 10. Scope of patent application: 1. A synchronous read from redundant method, the redundant RAID disk drive plus a single redundant disk drive, one of the multiple sectors is disassembled and disassembled Dimensions include steps from one to the capacity of the array: The data array of the Raid drive array has any number of data magnets, the stored data is one or more of the drives, and the read request is- Integer sector, this method pack Decide whether the array recognizes a first initial sector of the array; any drive has failed; a drive that contains read request data, reads data from the first drive, The place where the block is or goes to ^ ° Be the initial dagger i ,,., Which is requested, or it should be read out; sa &lt; ,,, °, no matter which comes first For the disk drive, read the magnetic field corresponding to the rest of the disk drive, and calculate a result of reading the data from the machine. ^ Wood X〇R function, and "immediately, transmission language! The method of the patent ... please include: if the reading outlet extends beyond a single continuous block of resources #, and read two # 出 出 了 数据。 The data from the disassembly until η cuck5 continuous disassembly and retrieval of these drives is the same as _ w out request is completed as m, all read access is called through the The method of item 2 of the data object further includes: if and if-... seek from-the failed disk drive, then read the magnetic range from the remaining range of the magnetic disk 85 200535609. @@ et — Gumen magnetic &amp;, a Brin XOP that reads the data from the remaining disk drives, Zhao α "R function, and real-time" transmission of the result. 4 · As stated in the patent scope The method of item 2 further includes: identifying the first disk drive, on which a data block is stored, which includes one of the final magnetic sectors of the read request data; and reading the second disk drive from the identification. Data; transmission of the second magnetic stele mother ancestor, & ^ is 钺 贝 钺, from the beginning of the block, or if it Tied to the same block and self-deprived from the private sector, such as the initial magnetic field until the final magnetic field; and if the final magnetic field F is from &lt; If one of the magnetic drives fails, it reads the remaining disks, and drags the magnetic disk corresponding to the consumption area, and calculates and reads out a Brin XOR function from the remaining disk drives of the Haihe Temple. And the result is transmitted as the data block of the final magnetic zone. 5. If the method of applying for the second item of the patent scope &lt; Wanfa, wherein the drive of the RAID array is a SAA disk;) #, Each of Gan ++, Yi ', and Qi is attached to a SATA port, and the synchronous data readout operation package is ^ ^ U step to transfer data from the controller side of each SATA port. F: two types for updating data blocks Method, the data block is stored in a decomposed structure at a corner with no money ^ ▼ the redundant disk array of the shell material block τ, δHai method includes steps: identifying a current data demolition Solution, one of which is the updated block; identify one of the arrays, the first magnetic a η ..., Zhu Qin 'its storage is used for the current dismantling purposes; parity-lean materials; 1 identify the — 筮-^ block of the array; I machine, which stores information to be updated 86 200535609 from the first And the second disk drive m μ + 出 out of this data block and read out a corresponding parity block; the first retrieved data block and one of the retrieved, pasted parity blocks are the first XOR to form an intermediate data block; store the calculated XOR in the storage location of _Temporary Eyes, where the diction step was "real time, and was made during the storage step ^ M times r less private period without the need Store the retrieved block and the parity block in memory; read the intermediate block and read the new data block; 'Judging the difference between the intermediate block and the new data area 埗, L The current second X〇R · Synchronous storage of the calculated second X 0R, R is stored in the same drive (the first disk stick) and the new data block is stored. in. 7. The method according to item 6 of the patent application, wherein the disk drive is connected via a serial port. 8. Method according to item 7 of the patent application. A disk drive connected via a SATA port. 9. The method according to item 8 of the scope of patent application is coupled to a buffer with lean path switching logic. 10. The method of item 9 in the scope of patent application The switching logic is reconfigurable. 11. If the switch logic system of item 9 of the scope of patent application includes: Boling X0R, Gallery M | 1 pezhu road catty is not a hard body, which is used to calculate the choice of the magnetic disk drive of Douji Temple Data input-ship. A two-step method for updating—a data block. The data block is a data structure that is disassembled and stored in a redundant block with the updated data block. The array system includes the array system. Including, the array is composed of the data path, the data path 87 200535609, and the method includes the steps: Fang, No. #I, read the data block from the array and read "Parity block" calculates the position of the data block and parity block, and stores the XOR side and the next day's storage location order without storing the data block or parity block in the memory; and One fun two steps' read out the calculated data from the temporary storage location and read the updated data block from memory, calculate the -calculated XOR and update data block ^ ^ ^ ^ Brother X0R, store The update data &amp; block is stored in the array, and the XOR of the 帛-巧 仔 巧 弟-Juyi is stored in the co-located drive to update the co-located block. Wherein, the array is I3. The method according to item 12 of the patent application includes: a disk drive connected via a serial port. Among them, the array is I4. The method is according to item 13 in the patent application: including a SATA. Grab the second, the drive connected by the μ Λ1Α port. Among them, the array is b. The method according to item 14 of the scope of patent application is coupled to the 'buffer' by the data path switch logic. % If the scope of the patent application is 15th: The path switch logic is inserted in the data path of the method, the method, and the method. ^ Y-The drive and the host's buffer memory are directly produced and accessed to synchronize: flush. . Directly enable the XOR calculation for 17.17 h at night and T on the same day. 2. AID disk arrays and multiple serial drives ^ 歹 is' contains: [write] "year-old" surface, used to attach a real disk array; connected to a shell disk drive In order to form each of these serial interfaces, ^ write data so that 1, 1 memory is used to store Ding 1 fixed on a magnetic disk ..., operation and stored on the attached disk drive 200535609; each of these serial interfaces provides-when the buffer memory is full; ~ -1 output to indicate the switching logic circuit, connected to the write of at least one direct memory access channel Person = 'used to provide the written data to these buffer records during the receiving period' and used in the write operation-logic circuit to measure when all: 'a control circuit, in response to the logic, σ fe body Is not full; _ Road 'is used to write synchronously from the direct memory access channel via the P3M and F »series circuits: Logic all buffer memories to form synchronous write data:; of And write-be expected to "The switch logic circuit includes, It is written from the synchronously written information column ^ i'P. Several data are stored for storage in the array. 18. For example, the controller of the item ρ of the patent application range, wherein the serial magnetic Computer interface "D disk array capacity interface. ^ The "disk lookup" system contains a "TA specification phase eight middle 4 controller is implemented on a daughter board. ° Control state is implemented on a computer motherboard. 2. As stated in the patent No. 18 of the improved Raid disk array manufacturing system, the switch logic circuit is configurable to implement a direct memory access channel and at least one SATA drive. Mediation expects relevance. 89 200535609 22. The controller of item 21 in the scope of patent application, wherein the switch logic voltage M ^ D disk array is configured as the mapping information in the scope of patent application. The controller, of which 9 is a modified RAID disk array ^ ^ /, ~ The mapping software is reconfigurable to configure the switch logic circuit. For the use of evil 24_ If the improved RAm controller of item 17 in the scope of patent application, its # disk array serial interface _ Ge = Off In single-disk write operation and write 5 · For example, the controller of item No. 丨 7 in the patent application scope of A, A6, good-quality RAID disk array (FIF0) memory. FIFO 17 Shi 3 2 1 ′ The switching logic circuit is based on a lean path that is based on a current configuration. The controller, January patented the improved RAID disk array of 26 items. The switch logic circuit is configured by mapping data. Control J. For example, please configure the improved one-disc array of the 27th item of the patent. The open-second software can be reconfigured for dynamic control of the 2 device. For example: She improved the 27th item. "Array" is implemented on a host bus adapter. Control 3 devices. For example: the improved magnetic disk array / in the 17th area of the patent, the switch logic circuit includes a disk write Accumulator 90 200535609 circuit. An improved RAiD disk array controller, including: a plurality of tandem disk drive interfaces, a disk drive array; &lt; contiguous disk drives to form a magnetic disk-1:歹] Each of them includes a 'buffer memory, which is used to store in the disk read operation # ^ 1 m H data; 纟 纟 亥 亥 read data captured by the attached disk drive Each of the serial interfaces # 裎 祝 ^ xiπ a # ^ is used for a status flag output to indicate whether the buffer memory is currently empty; ^ The switch logic circuit is pulled out and coupled to the path for receiving from These serial introductions = "to implement the data and provide it during a write operation. Output data ^ ^ ° Received read data to the at least one direct memory access channel; ^ a logic circuit for damage detection &amp; +, 饤 all buffer memory is not empty; a control circuit In response to this policy, η + will eventually be freely transmitted through the switch logic circuit to synchronize the transmission of all the material from the _ _ / &amp;, S, f-T ° to the direct The limb access channel is bad to form synchronous readout data, and the switch logic circuit includes Tai ... "... Brin X0R circuit, which is used to synchronize data from buffer memory such as Suan, etc." ”It ’s the turn of direct memory access channel%, that is, to reconstruct the missing data from the synchronization # Ψ 少 δ less than the missing data. 32. For example, the improved RAID disk display controller in the scope of the patent application, Among them, the tandem disk drive 'six series' includes a SATA-specific interface. 3 3 · If the patent application scope is λ-2 members of the improved RAID disk array 91 200535609 controller, in which the controller is implemented on a daughter board. 34. The improved magnetic disk array controller according to item 32 of the application, wherein the switch logic circuit is configurable, and is used for each direct memory access channel to be expected to be associated with at least one SAT ^ disk drive. <35. For example, the improvement of the ruler-gamma disk array manufacturing method of the 34th patent application scope, where 'the related logic circuit is by mapping * i6. For example, the improved RAID of the 35th patent application scope; ^ Column Mb 'where' The mapping data is software reconfigurable for kinetic energy configuration of the switch logic circuit. Heart 37 · If the improved r · controller of the 31st scope of the application for a patent application, wherein the bow industry array π ±, the switching logic circuit is in a single magnetic disk read and fall simultaneously with the read data will promote the redundant Yu information :: one or more of the middle channel. m want. Hidden body access 38. For example, the improved R · work system benefit of item 31 of the patent application scope, wherein each serial port buffer memory system includes (FIFO) memory. First-in, first-out, such as the modified form of the 31st item ... B system benefits, in which the 32-bit data path is implemented. The circuit is based on a current configuration 40. For example, the 39th controller in the scope of patent application, where the switch logic circuit # | ^ -type RAID disk array 4 1 4 is hit by β ... The error is configured by the mapping data 0: · Such wide range No. 4. Item H, wherein the mapping data is “reconfigurable for dynamic 92 200535609 to configure the switch logic circuit. 42. For example, the improved RAID disk array controller of item 31 of the patent application, which It is implemented in a host bus adapter. 43. The improved RAID disk array controller according to item 31 of the patent application, wherein the switch logic circuit includes a disk read accumulator circuit. Eleven, schema: as the next page