KR20130100727A - Shingle-written magnetic recording(smr) device with hybrid e-region - Google Patents

Abstract

PURPOSE: A singled write magnetic recording (SMR) device with a hybrid E-area is provided to be optimized by using tradeoff between performance and cost. CONSTITUTION: A singled write magnetic recording disk drive includes an I-area (13), a disk E-area (16), a memory E-area (19), and a system electronic device (21). The disk has a plurality of circular concentric tracks in which data is stored, and a magnetic thin film coating. The I-area includes the tracks of a first subset. The disk E-area includes the tracks of a second subset. The memory E-area includes an exception records that is stored in a non-volatility solid state memory. The system electronic device re-stages the exceptional record of first set from the disk E-area to the memory E-area according to the first order that is output from the disk E-area, and the set of exceptional records is recorded in the disk E-area according to the different sequence to the first order. [Reference numerals] (10) Storage device; (11) Host/user; (12) Thin films on a disc; (13) I-area; (14) Cache; (15) Guard area; (16) E-area; (19) E-area of a flash memory; (21) Electronic device of an SMR system

Description

Single write magnetic recording (SMR) device with hybrid E area {SHINGLE-WRITTEN MAGNETIC RECORDING (SMR) DEVICE WITH HYBRID E-REGION}

This application relates to provisional applications, filed May 23, 2011, and which is typically assigned the serial number 61 / 489,174. The benefit of this provisional application is 35 U.S.C. Claim in accordance with 119 (e).

The patent application, filed Jul. 18, 2011, commonly assigned serial number 13 / 135,953, incorporated by reference, describes SMR drive embodiments with the double write cache region referred to in this application.

FIELD OF THE INVENTION The present invention relates to the field of data storage architectures, in particular single-write magnetic recording (SMR) devices.

Conventional disk drives with magnetic media organize data on concentric tracks that are spaced apart. The concept of shingled writing is described as a method of increasing the area density of magnetic writing in the form of vertical magnetic writing. In a single write magnetic recording (SMR) medium, an area (band) of adjacent tracks is written to overlap one or more pre-written tracks. Tracks in a single structure must be written sequentially, unlike conventionally separated tracks, which can be written in any order. The tracks on the disc surface are organized into a plurality of single regions (also called I-regions). The direction of single writing for the I-region can be from inner diameter (ID) to outer diameter (OD) or from OD to ID. The disc can also be single in two directions on the same plane, where the two regions meet at approximately mid-diameter points. The number of tracks that are single structured together in one area is the main performance parameter of the single write. Once written in a single structure, individual tracks cannot be updated in place because they destroy the data by overwriting the overlapping tracks. Thus, from the user's point of view, single write data tracks sometimes consider similar additional dedicated logs. In order to improve the performance of SMR drives, part of the magnetic medium is allocated to one or more so-called "exception zones" (E-regions) that are ultimately used as staging areas for data to be written to the I-regions. E-regions are sometimes referred to as E caches. Most data in an SMR drive is expected to be stored sequentially in the I-areas, and data records not currently stored in the I-areas may be considered as an "exception" for sequential I-area storage. . When writes in random order have been received, they are generally stored in the disk E-area in the received order.

Sanvido et al. US Pat. No. 7,965,465 (June 21, 2011) discloses a cache memory to facilitate updating records in shingled blocks of SMR disk storage that must be written sequentially. Disclosed are techniques for use.

Address indirection in the internal architecture of a single write storage device is useful to protect the host from the complexity associated with SMR. Host file systems typically use logical block addresses (LBAs) of instructions for read and write blocks of data, ignoring the actual places (physical block address (PBA)) used internally by the storage device. do. Hard disk drives have, among other things, a level of LBA-PBA indirection for decades so that bad sectors on the disk can be remapped into good sectors reserved for this purpose. Address indirection is typically done in the controller portion of the architecture of the drive. This controller translates the LBAs in host commands into an internal physical address or an intermediate address from which the physical address can ultimately be derived.

Conventional LBA-PBA mapping for defects often does not need to be changed. In contrast, in the SMR apparatus, the physical block address PBA in the logical block address LBA may be changed frequently. The indirection system provides a dynamic translation layer between host LBAs and current physical locations on the medium. In an SMR system, LBA-PBA charting can change with every write operation because the system dynamically determines the physical location on the medium on which host data for the LBA is to be written. Data for the same LBA is written to another place the next time the host LBA is updated. Moreover, the drive automatically moves data between write caches in RAM, write caches on disk, E-regions on disk and I-regions on disk. LBAs for data remain intact, regardless of where the drive holds its stored data. Background processing, such as defragmentation, is also executed automatically by the device to move data sectors from one PBA to another while the LBA remains intact.

Fragmentation is generally a term often used to describe the process of reorganizing records in a file or database system to eliminate or reduce fragmentation. In the SMR I-region when records are updated or removed, the number of small free spaces, commonly referred to as invalid or 'stale' data, increases. The process of defragmentation makes these records closer by physically moving them, while creating larger and more useful free areas. DRAM is typically used for restaging because efficient sorting of records can be done in the proper sequence. For SMR drives, efficient defragmentation is an important factor in the overall performance of the device.

Embodiments of the present invention include SMR disk drives having a nonvolatile solid state memory E-region in addition to magnetic medium E-regions on the disk. The combination of solid state memory and E-regions on disk is referred to as a hybrid E-region. In embodiments of the present invention, the solid state memory E-region may be used for an operation referred to as destaging and / or restaging. The memory E-region can be used to sequence sets of exception writes to reduce the time and energy consumed when doing a search in the disk E-region. The ratio of the size of the solid state memory E-area to the total E-area capacity on the disk can be optimized for applications selected in accordance with the present invention using a tradeoff between performance and cost. For example, an embodiment having a memory E-area size that is 10% of the total disk E-area capacity, achieves substantially improved performance over the disk-only E-area example, and also achieves performance in the NAND-only memory E-area. This results in a reduction in the required cost.

In one embodiment, one set of incoming write exceptions are first stored in the solid state memory E-area and then reordered before being destaged in the disk E-area. By using the solid state memory E-area, reordering improves sequencing and reduces the searches required when data records are accommodated in disk storage.

In another embodiment, the non-volatile solid state memory E-region is used for restage of data that has been previously written to the disk that needs to be reordered, such as in a defragmentation process, and then rewritten to the disk. do. Destaging and restaging embodiments may be used together.

SMR disc drives are described that have a hybrid E-area on a disk that includes a non-volatile solid state memory E-area in addition to a magnetic medium E-area. The memory E-region may be used for operations referred to as destaging and / or restaging to sequence sets of exception writes to reduce time and energy consumption when performing searches in the disk E-region. The ratio of the size of the solid state memory E-area to the total E-area capacity on the disk can be optimized for the application selected according to the invention using the trade-off between performance and cost. For example, an embodiment having a memory E-area size that is 10% of the total disk E-area capacity achieves significant performance improvements over the disk-only E-area example and also requires less than that required in the NAND-only memory E-area. Enable cost.

1 illustrates an SMR data storage device having hybrid E-regions according to an embodiment of the present invention.
FIG. 2 is a graph showing sustained IOP on the vertical axis versus reread rate (seeks per second) on the horizontal axis, with an SMR drive with disk-only E-area, and NAND flash sizes of 1 GB, 2 GB, 4 GB and 8 GB. For four drives according to embodiments of the present invention having hybrid E-regions.
3 is a graph showing the E-region size required to meet PMR performance using NAND hybrid E-regions and disk-only E-regions for destaging in accordance with the present invention.
4 is a bar graph showing the ratio of user capacity to disk E-area size for devices with various NAND E-area sizes and DRAM reread buffer sizes and transfer lengths.
5 is a graph showing sequential read performance in MB / sec for randomly written regions for a conventional PMR drive, a disk-only E-region SMR drive, and a hybrid E-region drive according to the present invention. .
6 illustrates reordering of data blocks during destaging from a NAND flash E-region to a disk E-region according to one embodiment of the present invention.
7 illustrates the use of a NAND flash E-region when restage data blocks from a disk E-region to facilitate reordering of the data blocks during fragmentation removal of the disk E-region according to one embodiment of the present invention. Shows that.
8 illustrates destaging from the NAND flash E-region of the reordered data blocks back to the disk E-region during fragmentation removal of the disk E-region according to one embodiment of the present invention.

1 illustrates a data storage device (DSD) using SMR having a system electronic device 21 according to an embodiment of the present invention. The system electronic device 21 operates according to the prior art except as described herein, and aspects of the system for performing the conventional functions are not shown. The system electronic device 21 may be a conventional system-on chip, which is an integrated circuit including a host interface, a controller, servo functions, a microprocessor, a firmware program, and the like, on a single chip.

The host / user 11 can be any type of computer and can also communicate with the device by any means including a network. The term "user" will be used interchangeably with "host". Multiple hosts can also communicate with the device using conventional techniques. Thin films 12 are magnetic thin film coatings, which are typically deposited on both the top and bottom surfaces of a hard disk (not shown) and the device may have multiple disks. Membrane 12 is shown in cross section in FIG. 1. In plan view, the regions are a plurality of concentric circular bands. Magnetic thin films are formatted for use in an SMR architecture, in this embodiment the disk E-region 16, I-regions 13 (also referred to as I tracks), write cache regions 14 (doubled). Also referred to as write cache regions), and a guard region or bands 15. Although only one is shown in FIG. 1, the device may have multiple E-regions 16 on each disk surface and typically there are multiple disk surfaces. Non-volatile solid state memory E-region 19 is generally referred to as NAND flash E-region or NAND E-region. NAND flash E-regions 16 are used in accordance with the present invention to improve performance and save power over the example of disk-only E-regions. Moreover, cost is saved by designing the size of the NAND flash E-region 16 to be significantly smaller than that required for the NAND-only E-region.

Another analogy to the problem addressed by the present invention can be made for postal delivery. Incoming correspondences come in random order and need to be sorted into specific destinations such as country, region, state, zip code, postal route, delivery order. It would be very inefficient for a postman to deliver mail back and forth without seeing the address properly.

So in this analogy, write exceptions are received in basically random order, like letters, but writing them to disk in the order they come in can be very time consuming and resource-consuming when you later need to change the order of the exceptions. To the end. Exceptions need to be organized in several ways before they can be written efficiently to the I track of the I-region. High level line classification by postal code may be compared to organizing exceptions by target I-region. Other levels of classifying exceptions include LBA sequences. Categorizing the exceptions in accordance with the present invention enables maintenance write performance that is completely independent of the search speed since the number of searches is significantly reduced.

In addition to those described herein, general destaging and restaging algorithms may be used to handle data transfer back and forth between the NAND E-region and the disk E-region. In the following description, the nonvolatile solid state memory used in the E-region 19 is generally referred to as NAND flash because it is a preferred write buffer based on the current technology. However, any form of nonvolatile solid state buffer memory such as phase change memory, NOR flash, and / or MRAM may be used in accordance with the present invention. The use of the term "NAND" is for convenience of description and is not meant to limit the embodiments of the invention.

Destaging Using Hybrid E-Area

By using the NAND flash E-region 19 as a write buffer, one set of exceptions can be generally reordered in sequence when they are destaged into the disk E-region 16. It is desirable to accumulate a large set of exception records for sorting before writing to the disk E-area. Thus, destaging should generally be activated when there is no free space in flash E-region 19 to maximize the number of exceptions per I-region upon destaging. This reduces the number of seeks (rereads) when fragmentation is done.

6 is a diagram of reordering of exception data blocks during destaging from NAND flash E-region 19 to disk E-region 16 in accordance with one embodiment of the present invention. Over time, the host can send write records in sequential, reverse sequential, any other non-random order, or basically in random order. The present invention provides the greatest gain compared to the prior art when records are received in random order. In the example of FIG. 6, it is assumed that the left LBA blocks are received from the host / user in a random order by default and located in the NAND flash E-region. Although exception records can be sorted into sequences in the NAND flash E-area, the physical ordering of the exception LBA blocks within the flash E-area is important because mechanical and time-consuming searches are not required to move from input to input. Not. Importantly, the set of LBA blocks is destaged in the disk E-area in an improved sequential order. This means that disk E-region 16 does not contain completely random writes as in conventional SMR. Initial reordering of one set of host random writes (exception writes) increases the effective reread rate of individual exceptions. The number of searches is reduced, but this is somewhat independent of the reread rate.

These " skip-sequentials " are burst read out following restage / fragmentation removal The present invention allows for a substantial reduction in the number of random seeks during the defragmentation process for the disk E-region 16. Exception Recording When or before the set of these are written to the disk E-region 16, their reordering may be accomplished by firmware programs that are accomplished by the system electronics and also executed by the microprocessor.

Since more random rereads are needed in the disk-only E-region, this may include: a) a poor short search mechanism; b) high sorting overhead; c) insufficient SRAM for large internal queues; And d) high power consumption. In contrast, the present invention makes improvements to each of the above-mentioned problems with disc-only E-region design by allowing fewer random searches during rereading during restage / fragmentation removal, which saves considerable power. Make savings possible. Although DRAM can be used instead of NAND flash, using NAND rather than DRAM is a benefit. NAND is about 10 times cheaper per MB than DRAM, and NAND is nonvolatile unlike DRAM.

Hybrid E-region designs are also less expensive than replacing disk E-regions with 100% NAND flash E-regions. The present invention allows the size of the NAND E-region to be significantly smaller than the size of the entire disk E-region. For example, using the size of the NAND E-region as about 10% of the size of the total disk E-region storage as shown in the following data achieves improved performance. Using a 10% NAND E-region size also reduces the cost of increasing the NAND-only E-region design to 10%. Performance estimates (see Table 1) indicate that a system with 2 GB NAND used for destaging in a hybrid E-region design is comparable to a system with 32 GB NAND-only E-region (421 IOPs). 498 IOPs can be achieved. The estimates in Table 1 estimate 2TB user capacity and 4k RW IOPs. IOPs represent input / output operations per second.

NAND capacity (GB) % For save NAND E-Area IOPs NAND Destage IOPs 2 0.11% 30 498 4 0.21% 59 568 8 0.43% 116 626 32 1.72% 421 783 64 3.44% 747 954

Current vertical magnetic recording (PMR) drive technology places random writes at random locations, requiring more searching before each write and thus slowing response time. An SMR driver with a hybrid E-region has faster random write IOPs because random search is unnecessary while the user writes data. New user data is sequentially written to the I track of the I-area or the disk E-area of the SMR drive.

Figure 2 shows sustained IOPs on the vertical axis versus reread rate (seeks per second) on the horizontal axis, showing an SMR drive with disk-only E-area, and NAND flash sizes of 1 GB, 2 GB, 4 GB and 8 GB. Is a graph of four drives with hybrid E-regions with Each of the four hybrid E-region examples results in a faster IOP than the disk-only E-region over a reread rate range of 200 to 1000 seeks per second. Although larger NAND flash sizes result in better performance, doubling the size of each flash increases yield less than the previous one.

Sequencing the exceptions in accordance with the present invention almost completely eliminates the effect of search speed from maintenance write performance. NAND E-Area Destage (NED) achieves performance equivalent to NAND-only E-Area design and uses 80-90% less NAND. The same performance benefit is achieved by using equally sized DRAM E-region buffers instead of NAND, while NAND E-regions have advantages over DRAM.

The graph in FIG. 3 compares the long transport block size performance when using NAND hybrid E-regions and when using disk-only E-regions for destaging according to the present invention. The horizontal axis is the transmission length in 1 k units. Estimates are PMR performance level, command Queue Depth (QD) = 32, Write Cache Enabled (WCE) = 1, that is, WCE On. The vertical axis is the ratio of the percentage of disk E-area size to user capacity. If the size of the disk E-area is allowed to stretch, the reread rate has a strong impact on the amount of E-area space required to meet the PMR equivalent performance level. E-region stretching causes reread to be dominant. For a given exception reread rate, the E-region size requirement is doubled every two times the transmission length. This keeps the average number of exceptions per track constant.

The upper three lines in the graph are for the reread rate of 200, 500 and 1000 searches per second for the disk-only E-region. The lower line is for the hybrid E-region according to the invention. In accordance with the above assumption, with a transmission length of 1 k and a reread rate of 200, the present invention allows the PMR performance to be met with the size of the disk E-region being about 10% of the size of the required disk-only E-region. The benefits of hybrid E-regions decrease with increasing transmission length, but they are never worse than disk-only E-regions. For drives with low or reread rates, NAND destages (“NEDs”) can reduce the amount of disk E-area required to allow more drive capacity to be allocated to the I-area for user data storage. Can be used to reduce. (Note: Low reread rates for one drive versus another can result from a design tradeoff, for example, to save space, minimize power to seek, or lower costs. Actuator An example would be the selection of smaller and less powerful voice coils that serve to move.

Restage with Hybrid E-Area

NAND flash E-region 19 may also be used for restage (e.g., fragmentation removal of I-region or disk E-region) as well as initial reordering (sequencing) prior to destaging, as described above. . Figure 7 shows a NAND flash E-region when restage data blocks from disk E-region 16 to facilitate reordering of data blocks during fragmentation removal of the disk E-region according to one embodiment of the present invention. It is an illustration of using (19). 8 is an exemplary diagram of destageting of reordered data blocks from a NAND flash E-region back to the disk E-region during fragmentation removal of the disk E-region according to one embodiment of the present invention. This restaging may be accomplished by system electronics having a firmware program executed by the microprocessor. In the defragmentation operation of the disk E-area, by using the flash E-area, records are read from the disk E-area in the physical order in which they appear to minimize seek time. Once the records are in the flash E-region, they can be easily sorted using any selected criteria, such as the LBA sequence.

The benefits of restage using NAND E-regions include:

예컨대 DRAM의 16MB 내지 64MB로, DRAM 필요량을 감소시킨다 (DRAM은 NAND 플래시보다 10배 이상 고가이다). DRAM 버퍼 크기는 디스크 E-영역 크기의 필요량에 큰 역할을 수행한다.

For example, 16 MB to 64 MB of DRAM, which reduces DRAM requirements (DRAM is 10 times more expensive than NAND flash). The DRAM buffer size plays a large role in the required amount of disk E-region size.

리스테이징은 I 트랙 재기입에 앞서 예외들을 재순서화하기 위한 부가적 기회이다.

Restage is an additional opportunity to reorder exceptions prior to I track rewriting.

The bars in the graph in FIG. 4 represent the ratio of the percentage of disk E-area size to user capacity of devices with various parameters. Three bars 3 of each set correspond to the selected NAND E-region and DRAM reread buffer size shown below the bar. For example, the first set of bars on the left is for NAND = 2GB and DRAM = 32MB. For each set of bars, from left to right, the bars correspond to transmission lengths of 128k, 64k and 32k. This result shows that the DRAM buffer size plays a large role in the disk E-region size requirement. Thus, it may be beneficial to write exceptions to the NAND E-region immediately before the defragmentation operation. For example, small DRAM can be included in the design by using the NAND E-region instead of DRAM for restage.

If the disk E-area is much larger than the available NAND for destaging, it may be beneficial to save the exceptions back to the NAND just prior to defragmentation by restage to the NAND. Widespread use of DRAM can also be integrated with the I-track. For example, if the E-region is ten times the size of the NAND destage buffer, an average of ten seeks will be needed to recover all exceptions from the disk E-region for the I track prior to the defragmentation operation. Multiple I tracks of exceptions must be reread for each seek, which increases the buffer requirement and reduces the reread cost per I track.

The graph of FIG. 5 shows sequential read performance in MB / sec for randomly written regions for conventional PMR drives, disk-only E-region SMR drives and hybrid E-region drives according to the present invention. The horizontal axis represents the transmission length. This result reduces the number of random reads required for NAND E-region destaging according to the present invention to satisfy sequential reads in a host region containing randomly written data compared to a disk-only E-region SMR drive. Show what you're asking

Alternative Embodiments and Optional Features

Some optional features and optimizations that may be included in various embodiments of the present invention are described below. One option is to position the disk E-region 16 near the OD of the disk to use the high bits per revolution that the hard drive has in the OD. The tracks near the OD keep about twice as much data per revolution at the longer, typical disc ID.

Another option is to select full disk E-area storage as about 3% of total disk storage, such as 30 GB in a 1 TB SMR drive. Since disc E-area tracks are used instead of expensive I-area tracks, the advantage of the present invention is that the disc E-area size can be made smaller than in other ways. The use of larger NAND storage capacity weakens the growing benefit of larger E-region capacity (see Figure 2).

As mentioned above, selecting the total NAND E-area storage to be approximately 10% of the total disk E-area storage or 0.3% of the total disk storage is an attractive option for achieving significant performance improvements at a reasonable cost. If the total disk E-area storage is selected as about 3% of the total disk storage and the NAND E-area is selected as its 10%, the NAND E-region is 0.3% of the total disk storage.

Another option to consider is that the DRAM can be used for buffering of incoming write operations before writing to either NAND and / or disk; DRAM and SRAM are used by hard disk controllers to help reorder and move user data.

Claims (13)

A disk having a magnetic thin film coating, wherein data is stored in a plurality of circular concentric tracks;
An I-region comprising tracks of a first single written first subset having tracks partially overlapping previous tracks;
A disk E-area containing tracks of a second subset of holding exception records;
A memory E-region containing exception records stored in non-volatile solid state memory; And
Restage a set of exception records from the disk E-area to the memory E-area in a first order read out from the disk E-area, and then set the set of exception records in a different sequence order than the first order. A single-structure magnetic recording disc drive comprising system electronics writing to an area. The method of claim 1,
And a plurality of disk E-areas having a total disk E-area storage size, wherein the size of the memory E-area is substantially smaller than the total disk E-area storage size. 3. The magnetic recording disk drive of claim 2, wherein the size of the memory E-area is about 10% of the total disk E-area storage size. 3. The unitary structure of claim 2, wherein the sequence ordering takes less searching time needed to write the sets of exception records than is needed to write the sets of exception records in the read first order. Magnetic recording disk drive. 2. The method of claim 1, wherein the sequence ordering takes less searching time needed to write sets of exception records into the I-area than is required to write sets of exception records in the first order read out. Magnetic recording disk drive, single structure. The method of claim 1,
The system electronics stores a second set of exception records received with write commands in a received order in a memory E-region, and then stores the second set of exception records in a different sequence order than the received order. A single-structure magnetic recording disk drive that writes to the disk E-area. Reading a set of exception records from the disk E-area to the memory E-area in the non-volatile solid state memory, the set of exception records being in a first order corresponding to the physical order in the disk E-area. Read out; And
Writing the set of exception records in a disc E-region in a different sequence order than the first order, wherein the sequence order is a sorting order for a selected characteristic of the set of exception records. How a magnetic recording disk drive works. 8. The method of claim 7, wherein the sequence ordering takes less searching time needed to write sets of exception records into the I-area than is required to write sets of exception records in the first order read out. Operating method of magnetic recording disk drive of single structure. 9. The disk drive of claim 8, wherein the disk drive further comprises a plurality of disk E-areas having a total disk E-area storage size, wherein the size of the memory E-area is substantially smaller than the total disk E-area storage size. Method of operation of single structure magnetic recording disk drive. 10. The method of claim 9, wherein the size of the memory E-area is about 10% of the total disk E-area storage size. 11. The method of claim 10, wherein the sequence ordering takes less searching time needed to write the sets of exception records into the I-area than is needed to write the sets of exception records in the read first order. Operating method of magnetic recording disk drive of single structure. 8. The unitary structure of claim 7, further comprising the step of sorting the set of exception records in sequence order in the memory E-region before the writing of the sets of exception records in the disk E-region in sequence order. How a magnetic recording disk drive works. The method of claim 7, wherein
Storing a second set of exception records in a memory E-area, wherein the second set of exception records are received as write commands in a random order; And
Writing a second set of exception records to the disc E-area in a second sequence order that is different from the random order, wherein the second sequence order is a sorting order for a selected characteristic of the second set of exception records. A method of operating a magnetic recording disk drive having a single structure.

Images