CN105302472A - Operating method for distributed memory disk cluster storage system - Google Patents

Abstract

The invention relates to an operating method for a distributed memory disk cluster storage system. According to the method, a distributed memory storage system is used; expansion of four types of requirements can be met, i.e. expansion of network bandwidth transmission, expansion of disk capacity, expansion of an IOPS speed, and expansion of a memory I/O transmission speed; meanwhile, cross-region, cross-data-center and cross-WAN use can be achieved, so that a service can be provided in a centralized manner in local clustered memory disks according to user demands, and the capacity of the clustered memory disks also can be progressively expanded, so that a cross-region or transnational data service is further provided.

Description

Distributed memory disk cluster stocking system How It Works

Technical field

The networking information interchange stocking system that one possesses the multiple features such as the transmission of quick multi-to-multi, high extendibility and stability, about a kind of distributed memory disk cluster stocking system How It Works, refers in particular in system of the present invention.

Background technology

In in recent years, along with the fast universal of networking application and the networking demand growth of running gear, the very large change that allowed enterprise's message system have.The demand of counter resource never reaches height so, under the customer volume in the face of exploding with the time, is line and the bandwidth demand being unable to cope with such vast number user with current storage facilities.

First, in the processing server hardware system of networking of the prior art, electronic information need through networking bridge joint as the mutual approach to biography each other, but when quantity of information is positioned at normal level, although network transmission amount is enough to deal with, but when quantity of information is uprushed, because the transfer rate of networking bridge joint transmission has its higher limit, therefore no matter the hardware grade of server computer own is why, also the processing speed of bulk information cannot be increased, and in the face of message delay or interruption etc. awkward situation when can make public use network system, this is the disappearance of a current technology.

In addition, in prior art, because the technical development of internal memory is not extremely now highly, the therefore limited event of its storage capacity, only can as temporary use, therefore the server host hardware specification of tradition is all with the configuration space of hard disc apparatus as main storage and main operation system.

And the information transmission between hardware and hardware, the information of processing unit and internal memory to the speed of biography far away higher than the transmission speed between processing unit and hard disc apparatus, that is, be that major storage space is to do the support of computing with hard disc apparatus in known techniques, one cause treatment efficiency cannot reach the main cause of the best, especially when process information amount is large, this problem highlights amplification especially, in addition more leisure opinion hard disc apparatus especially far away can not compared with internal memory on serviceable life, through hard disc apparatus as main storing mode for total system, the guarantee upper in application is limited, non-is the pattern an of the best.

In summary, during the information that the process device processes of networking information is huge in prior art, its usefulness cannot play the problem of the true treatment effeciency of processor can See one spot, therefore described in before taking a broad view of, the inventor of the present invention thinks and designs a kind of distributed memory disk cluster stocking system How It Works, improved to the disappearance for prior art, and then the enforcement in enhancement industry utilizes.

Summary of the invention

Because aforementioned the deficiencies in the prior art point, system of the present invention design one possesses the distributed memory disk cluster stocking system How It Works of the patent requirements such as novelty, progressive and industry applications, to overcoming the difficult point of prior art.

For achieving the above object, the technology used in the present invention means are a kind of distributed memory disk cluster stocking system How It Works of design, it is: first do building of dispersion memory type storage facilities, and it is comprise the computer unit of plural platform composition cluster architecture and form memory disk of trooping; Then in computer unit installation system virtual machine platform operating system, installing by this makes computer unit form multiple virtual machine, and the mode that the setting of this computer unit utilizes operating system or program software to carry out set memory capacity to take, this internal memory is formulated for storage device, makes plurality of blocks memory disk; And then each archives is splitted into the data that arrives many parts, keeping the duplicate of to many parts to copy average mark is dispersed on all block memory disks, and use multi channel internal memory bus-bar parallel access memory modules, and become disk to use the capacity planning of memory modules, and wherein all operations system file form is supported in the access of this memory modules, and adopt and distributedly deposit framework, can be more than duplicate copies portion by data; When the operating system of virtual machine directly accesses by CPU the archives needed in memory modules, deposit back after process data in memory modules, virtual machine does the internal memory of calculating use also in memory modules, calculated the memory modules position of still directly depositing go back in deposited VME operating system, to allow in this stocking system data can in computer unit in the mode of parallel calculation in block memory disk process data; Other not the data system of computer unit through networking link connect a Port group now with chain to access the block memory disk of other computer units; And each computer unit is further according to functional one first data center, at least one second data center and the backup center divided into; Wherein feel concerned about with a virtual cluster data master station keyholed back plate in this first data backup, wherein respectively this second data center is with a virtual cluster data backup station keyholed back plate, and wherein this backup center system is with virtual cluster data backup station keyholed back plate, and wherein this first data center and this second data center form a distributed memory file system jointly; In addition, further for forming stack construction expansion storage volume framework, it uses the access mode of networking layer interface block memory disk multiple in a computer unit to be formulated for the memory disk element resources pond of trooping of a concept of trooping, and all block memory disks of synchronous use obtain data; After cluster architecture produces, each cluster architecture can both independently use, and each cluster architecture can be used as dispersion memory type storage facilities unit, use networking layer interface storehouse simultaneously, each dispersion memory type storage facilities unit can be emulated into large-scale memory disk, and the treatment capacity adopting new troop data master station and data backup station of trooping to carry out control data is dispersed in all large-scale memory disks.

Wherein, respectively this block memory disk system is electrically connected at least one hard disk memory device respectively, and this hard disk memory device system can the data an interval time by this block memory disk backup.

Wherein, the block memory disk in all computer units, all can use continuous data protected mode, is unremittingly backed up in unified large-scale physical hard disk cluster disk array by data backup continuously.

Wherein, this computer unit system has had the calculator device of CPU, internal memory, hard disk, networking card, mainboard, I/O adapter, connecting line, shell.

Wherein, every part of duplicate can both adopt 1-4096 position AES and RSA to use encrypted data with to leave on internal memory, whenever will accessing data, data flows between internal memory and CPU, virtual machine system is present in memory modules with archives form, virtual memory plan that the memory size of use is also at same block.

Wherein, respectively this block memory disk does the monitoring of its function situation through a monitoring unit, this detecting unit utilizes the software program of Splunk or other search engines to monitor it, more can configure when a problem is detected and restart application program service, reach the function of a reply.

Wherein, the version after virtual machine platform operating system employing VMwarevSphereESXi4.1, the version after MicrosoftServer2012R2Hyper-V, CitrixXenServerOracleVM, OeacleVM, RedHatKVM or Linux-BaseKVM.Wherein VME operating system adopts Linux, FreeBSD, OpenSolaris or MacOSX.

Wherein, be use storage area networking framework to run between the internal memory of virtual machine, take software virtual networking layer interface out, link together all block memory disks Collaboration.

Wherein, this networking layer interface, system's use SAN, SANiSCSI, SANFC, SANFCoE, NFS, NAS, JBOD, CIFS or FUSE interface carries out the communication between server and disk drive equipment, and exploitation has RAMSTORAGEAPI to use as application program support.Wherein this RAMSTORAGEAPI system adopts the RAMSTORAGE API that REST, Restful, C++, PHP, Python, Java, Perl or Javascript program development software (PDS) is formed; access dispersion memory type storage facilities unit AP I function, it comprises fault-tolerant, backup, transfer, fast arrangement virtual machine, hyperdisk size, automatically automatically increases according to actual demand capacity the function, continuous data protection, snapshot, monitoring resource etc. that between block memory disk, block, the transfer of data load balance, backup are replied.

Wherein, the resource of untapped CPU, internal memory, actual physics hard disk, system uses virtual machine platform operating system that all MRP are become unified resource pool, makes computer resource required for each, can adjust and be passed to the more abundant computer unit of other resources.

Wherein, the connected mode of multiple dispersion memory type storage facilities unit is according to the Internet protocol transmission of entity, and adopt SSLVPN, VPN or encryption calculation mode to transmit package, when networking line is obstructed, each region can both operate separately, when line is replied, data by complete synchronization to each block memory disk in each dispersion memory type storage facilities unit.

Wherein, this CPU is x86, x86-64, IA-64, Alpha, ARM, SPARC32and64, PowerPC, MIPS, Tilera Shu kind processor wherein.

Wherein, the internal memory use-pattern of this computer unit be directly by CPU Memory Controller Hub directly or hyperchannel many with threeway with speed direct access internal document more than 800MHz to 1333MHz.

Wherein, memory size is 1MB to 16ZB.Wherein adopted internal memory kind is dynamic asynchronous memory dram, dynamic synchronization memory dram, dynamic mobile platform internal memory, DYNAMIC DRAWING process internal memory, dynamically Rambus internal memory, static RAM, only rdma read or flash memory.

Wherein, dynamic asynchronous memory dram is FPMRAM or EDORAM.Wherein dynamic synchronization memory dram is SDRAM, DDRSDRAM, DDR2SDRAM, DDR3SDRAM, DDR4SDRAM or DDR5SDRAM.

Wherein dynamic mobile platform internal memory is LPDDR, LPDDR2, LPDDR3 or LPDDR4.Wherein save as VRAM, WRAM, MDRAM, SGRAM, SDRAM, GDDR, GDDR2, GDDR3, GDDR4, GDDR5, GDDR6 or GDDR7 in DYNAMIC DRAWING process.Wherein save as RDRAM, XDRDRAM or XDR2DRAM in dynamic Rambus.Wherein flash memory is NORFlash, NANDFlash, 3DNANDFlash, V-Flash, SLC, MLC, eMMC or TLC.

Wherein, hard disk memory device is traditional magnetic head, the hard disk of disk, solid state hard disc, networking hard disk, SAS interface hard disk, SATA interface hard disk, mSATA interface hard disk, mPCIE interface hard disk, FC interface hard disk, scsi interface hard disk, ata interface hard disk, NANDFlash add-on card or FCoE interface hard disk.

Wherein, networking card is Etherloop, Fast Ethernet road, gigabit Ethernet road, optical fiber, token-ring network, InfiniBand, FCoE or Wi-Fi.

Wherein, networking speed adopts the networking speed of 2Mbit/s, 10Mbit/s, 11Mbit/s, 40Mbit/s, 54Mbit/s, 80Mbit/s, 100Mbit/s, 150Mbit/s, 300Mbit/s, 433Mbit/s, 1000Mbit/s, 1Gbit/s, 8Gbit/s, 10Gbit/s, 16Gbit/s, 32Gbit/s, 40Gbit/s, 56Gbit/s, 100Gbit/s, 160Gbit/s or more than 1000Gbit/s.

Wherein, any one processor in mainboard compatible x86, x86-64, IA-64, Alpha, ARM, SPARC32and64, PowerPC, MIPS, Tilera.

Wherein, operating system file form is AdvFS, BFS, Btrfs, Coda, CrossDOS, DFS, Episode, EFS, exFAT, ext, FAT, GFS, HFS, HFSPlus, HPFS (High-Performance File System), the general parallel file system of IBM, JFS, Macintosh File System, MINIX, NetWare file system, NILFS, Nove11 store-service, NTFS, QFS, QNX4FS, ReiserFS, SpadFS, UBIFS, Unix file system, VeritasVxFS, VFAT, WAFL, XFS, Xsan, ZFS, CHFS, FFS2, F2FS, JFFS, JFFS2, LogFS, NVFS, YAFFS, UBIFS, DCE/DFS, MFS, CXFS, GFS2, Google file system, OCFS, OCFS2, QFS, Xsan, AFS, OpenAFS, AFP, MS-DFS, GPFS, Lustre, NCP, NFS, POHMELFS, Hadoop, HAMMER, SMB (CIFS), cramfs, FUSE, SquashFS, UMSDOS, UnionFS, configfs, devfs, procfs, specfs, sysfs, tmpfs, WinFS, EncFS, EFS, ZFS, RAW, ASM, LVM, SFS, MPFS or MGFS.

Beneficial effect of the present invention can be summarized as follows: the distributed memory disk cluster stocking system How It Works of the present invention is upper in design is through under system architecture method proposed by the invention, we adopt distributed memory stocking system, the expansion of four kinds of demands can be met, i.e. the expansion that the expansion of the expansion of networking bandwidth for transmission, disk size, the expansion of IOPS speed, internal memory I/O transmission speed increase.Can accomplish to meet the use of trans-regional, data center, WAN simultaneously, user's request can be concentrated at the memory disk of trooping (MemoryDiskCluster) of locality service is provided, also can gradual expansion to troop memory disk capacity, trans-regional or transnational Information Service is provided further.

In the increase along with reservoir, often increase a station server just networking bandwidth of a cumulative machine and disk size, as formation one resource pool (ResourcePool), this dispersion memory type storage facilities unit is just as an entity hard disk, wherein overall operation can not be affected during an entity host fault, in duplicate block memory disk can again backup data to new block memory disk, maintain certain data backup basis, also take continuous data to copy (ContinuousDataProtector simultaneously, CDP) for the backup of enterprise's data and reply provide innovative technology, to abolish tape backup often failure, a problem within one day, can only be backed up.

Also allow duplicate generation data can be sent by different blocks memory disk and reach many-one data transmission, when number of users increases, as long as increase block memory disk quantity, the transmission of multi-to-multi can be reached, most hard-disk crashes of RAID are allowed to cause the shortcoming of overall data loss, the crowded delay caused of data that the restriction of reservoir network interface quantity and speed causes user too much, LUN expands the problem of quantity, data center cannot the problem of transregional use, because adopt memory type disk system, each archives or each virtual machine can both be present in internal memory by archives form, the I/O high speed that direct use internal memory bus-bar is the highest, data is moved mobile in CPU and internal memory, I/O number, distance, speed is all the highest, take a broad view of above-mentioned advantage, be for known techniques institute can not and person, therefore can its incremental place of See.

And in order to allow above-mentioned purpose, technical characteristic and actual implement rear incrementally more to become apparent, below by be preferably implement example assist a ruler in governing a country corresponding be correlated with graphicly carry out more detailed explanation.

Accompanying drawing explanation

Fig. 1 is the enforcement schematic diagram of the distributed memory disk cluster stocking system How It Works of the present invention.

Fig. 2 is also the enforcement schematic diagram of the distributed memory disk cluster stocking system How It Works of the present invention.

Wherein, cluster architecture (1), dispersion memory type storage facilities unit (1A), computer unit (10), first data center (101), data of trooping master station (1011), second data center (102), to troop data backup station (1021), backup center (103), block memory disk (11), hard disk memory device (12), networking card (13), Port group now uses chain (20), distributed memory file system (40).

Embodiment

For effect that the expensive auditor of profit understands the inventive features of the present invention, content and advantage and can reach, hereby the present invention is coordinated accompanying drawing, and be described in detail as follows with the expression-form of embodiment, and wherein use graphic, its purport is only the use of signal and aid illustration book, may not be actual proportions after the invention process and precisely configure, therefore should with regard to appended graphic ratio and configuration relation is not understood, the interest field limited the invention in actual enforcement.

The advantage of the present invention, feature and the technical method reached are explained in more detail with reference to exemplary embodiments and institute's accompanying drawings and are easier to understand, and the present invention or can realize in different forms, therefore should not be understood the embodiment being only limitted to the statement of this place, on the contrary, have art and usually know the knowledgeable, the embodiment provided will make this exposure more thorough with comprehensively and intactly pass on category of the present invention, and the present invention will only be attached claim defined.

And unless otherwise defined, allly be used in term (comprising science and technology and scientific terminology) hereinafter and proper noun, in being in fact generally understood with the technology personage in this field belonging to the present invention equivalent in meaning, and those terms such as defined in general used dictionary should be understood to have the meaning consistent with the content of association area, unless and be defined in hereinafter significantly, to not understand with the excessively idealized or excessively formal meaning, conjunction is first chatted bright.

Shown in Fig. 1, the present invention proposes a kind of distributed memory disk cluster stocking system How It Works, wherein this distributed memory disk cluster stocking system How It Works in one preferably embodiment can be:

First do building of dispersion memory type storage facilities, it is comprise the computer unit of plural platform (10) composition cluster architecture (1) to form memory disk of trooping; And wherein this computer unit (10) is the calculator device of CPU, internal memory, hard disk, networking card, mainboard, I/O adapter, connecting line, shell of having had.

Main in computer unit (10) installation system virtual machine platform operating system, installing by this makes computer unit (10) form multiple virtual machine, and this computer unit (10) setting need use machine memory source capacity, the mode that system utilizes operating system to carry out set memory capacity to take, or utilize program software that this internal memory is formulated for hard disk unit, make a block memory disk (chunkmemorydisk) (11), this is just as the magnetic track of hard disk.

And then be able to a data archives being splitted into many parts, archives size or can be more than 64MB, keeping the duplicate of to many parts to copy average mark is dispersed on all block memory disks (11), therefore data is actual leaves in memory modules, and use multi channel internal memory bus-bar parallel access memory modules, and become disk to use the capacity planning of memory modules, and wherein all operations system file form is supported in the access of this memory modules, and adopt and distributedly deposit framework, can be more than duplicate copies portion by data, mode by this, even if machine damage is damaged with data center, data center still can operate.

Every part of duplicate data can adopt 1-4096 position AES and RSA to use encrypted data with to leave on internal memory, whenever will accessing data, data flows between internal memory and CPU, spent I/O access times and distance are minimum, virtual machine system is present in memory modules with archives form, virtual memory plan that the memory size of use is also at same block.

When the operating system of virtual machine directly accesses by CPU the archives needed in memory modules, deposit back after process data in memory modules, virtual machine does the internal memory of calculating use also in memory modules, calculate the memory modules position of still directly depositing go back in deposited VME operating system, relatively according to the I/O speed that the minimizing of access path is the fastest with use, allow this stocking system major part data in computer unit (10), data can be processed in the mode of parallel calculation at block memory disk (11), other just do not connect a Port group through networking card (13) in the data of computer unit (10) and now use chain (20) to access the block memory disk (11) of other computer units (10).

Wherein, virtual machine platform operating system, the version after can VMwarevSphereESXi4.1 being adopted, the version after MicrosoftServer2012R2Hyper-V, CitrixXenServerOracleVM, OeacleVM, RedHatKVM, Linux-BaseKVM; And VME operating system, can adopt and can adopt Linux (Linux2.6.14anduphaveFUSEsupportincludedintheofficialkern el), FreeBSD, OpenSolaris, MacOSX.

Moreover, respectively this block memory disk (11) more may do the monitoring of its function situation through a monitoring unit, this detecting unit maybe can utilize the software program of Splunk or other search engines to monitor it, more can configure when a problem is detected and restart application program service, reach the function of a reply, and the program of this type is known techniques, therefore further do not add to repeat in this part.

And each computer unit (10) is further according to functional one first data center (101), at least one second data center (102) and the backup center (103) divided into; Wherein this first data center (101) is with virtual cluster data master station (1011) keyholed back plate, wherein respectively this second data center (102) is with virtual cluster data backup station (1021) keyholed back plate, and wherein this backup center (103) is with virtual cluster data backup station (1021) keyholed back plate, and wherein this first data center (101) and this second data center (102) form a distributed memory file system (40) jointly.

And please coordinate referring to shown in Fig. 2 further, further wherein for forming stack construction expansion storage volume framework, it uses the access mode of networking layer interface block memory disk (11) multiple in a computer unit (10) to be formulated for the memory disk element resources pond of trooping of a concept of trooping, and its operation principles is as the bus-bar use-pattern of computing machine.

When adopting the block memory disk (11) of in multiple stage computer unit (10) 64, just use as the CPU bus-bar form of 64, the all block memory disks (11) of synchronous use obtain data, and when upwards expanding block memory disk (11) quantity, 128 or 256 are promoted to as by the CPU bus-bar of 64, the speed of its access promotes in cumulative mode, therefore wherein memory disk capacity also can increase because of block memory disk (11), the disk size upper limit also can increase, and promote the fiduciary level of data access speed and data, user can according to the gradual increase of demand.

When cluster architecture (1) produces, each cluster architecture can independently use, and each cluster architecture (1) can be used as dispersion memory type storage facilities unit (DMDCSDistributedMemoryDiskClusterStorage) (1A), use networking layer interface storehouse simultaneously, each dispersion memory type storage facilities unit (1A) can be emulated into large-scale memory disk (chunkmemorydisk), the treatment capacity adopting new troop data master station (1011) and data backup station (1021) of trooping to carry out control data is dispersed in all large-scale memory disks (chunkmemorydisk).

This just uses the resource of each main frame as parallel calculation, data is divided into block and delivers to each machine arithmetic, is finally combined into a result.

One of them block memory disk (11) or one of them dispersion memory type storage facilities unit (1A) were lost efficacy, and all can not affect the use of whole disk or make whole disk crash.

In addition, internal storage location fault in one one computer unit (10), the internal memory damaged can be indicated fault by this server computer unit, and does not re-use the block internal memory IC in this DIMM internal memory, when after replacing internal memory, the use of this resource just can be recovered.

And be use storage area networking (StorageAreaNetwork, SAN) framework to run between the internal memory of wherein virtual machine, take software virtual networking layer interface out, link together all block memory disks Collaboration; And this networking layer interface, system's use SAN, SANiSCSI, SANFC, SANFCoE, NFS, NAS, JBOD, CIFS, FUSE interface carries out the communication between server and disk drive equipment, and exploitation has RAMSTORAGE API to use as application program support.

Again wherein, this RAMSTORAGE API system adopts REST, Restful, C++, PHP, Python, Java, Perl, the RAMSTORAGEAPI that Javascript and other program development software (PDS)s are formed, access dispersion memory type storage facilities unit (DMDCSDistributedMemoryDiskClusterStorage) (1A) API function, it comprises fault-tolerant, backup, transfer, quick arrangement virtual machine, hyperdisk size, automatically block memory disk (11) is automatically increased according to actual demand capacity, data load balance transfer between block, the function that backup is replied, continuous data protection (CDP), snapshot, monitoring resource etc.

In addition respectively this block memory disk (11) is be electrically connected at least one hard disk memory device (12) respectively, this hard disk memory device (12) is the data an interval time by this block memory disk (11) to backup, produce to prevent unexpected fault, such as, by modified data, the part actual copy of difference portion can be arrived actual physics hard disk memory device (12) backup information in each block memory disk (11) each minute.

After each block memory disk (11) restarts, capital will finally be backed up in the upper data of actual physics hard disk memory device (12), complete reverts on block memory disk (11), and notice is trooped, data master station (1011) adds cluster environment running.

Block memory disks (11) all in all computer units (10), capital uses continuous data protected mode (ContinuousDataProtector, CDP), unremittingly continuously data backup to be backed up in unified large-scale physical hard disk cluster disk array, whenever part server computer unit or Partial Block memory disk (11) were because environment or other factors lost efficacy, snapshot or time point reduction virtual machine can be found out according to the time needed, and wherein said large-scale physical hard disk is aforementioned backup center (103), certain cluster disk array also can make the 3rd part of full backup by traditional tape mode.

Each block memory disk (11) can use the mode of automatic subordinate to plan required capacity and cpu resource, networking layer interface also can set required networking IP and MACaddress automatically, and a kind of actual demand setting AP on a virtual machine, and impose a condition needed for can automatically assigning.

In addition, if the resource of untapped CPU, internal memory, actual physics hard disk, can use virtual machine platform operating system that all MRP are become unified resource pool, computer resource required for each, automatically can adjust and be passed to the more abundant computer unit of other resources (10).

The connected mode of multiple dispersion memory type storage facilities unit (1A) then can according to the Internet protocol transmission of entity, and adopt SSLVPN, VPN or encryption calculation mode to transmit package, and mode that is trans-regional, national, WANIP can be reached operate, when networking line is obstructed, each region can operate separately.When line is replied, data by complete synchronization to each block memory disk (11) in each dispersion memory type storage facilities unit (1A).

And during enforcement aspect wherein in some hardware specifications selects, this CPU is the one kind of processor of x86, x86-64, IA-64, Alpha, ARM, SPARC32and64, PowerPC, MIPS, Tilera.

The internal memory use-pattern of this computer unit (10) be directly by CPU Memory Controller Hub directly or hyperchannel many with threeway with speed direct access internal document more than 800MHz to 1333MHz.

The single internal memory capacity wherein used is 1MB (megabyte) to 16ZB (zettabyte), and the internal memory kind adopted can be enumerated form for dynamic asynchronous memory dram and be: FPMRAM, EDORAM; Or dynamic synchronization memory dram enumerate form for: SDRAM, DDRSDRAM, DDR2SDRAM, DDR3SDRAM, DDR4SDRAM, DDR5SDRAM and what is called list model by increasing the new upward-compatible kenel model of different from access mode generation according to access speed; Or dynamic mobile platform internal memory enumerate form for: LPDDR, LPDDR2, LPDDR3, LPDDR4 and what is called list model by increasing the new upward-compatible kenel model of different from access mode generation according to access speed; Or DYNAMIC DRAWING process internal memory is enumerated form for VRAM, WRAM, MDRAM, SGRAM, SDRAM, GDDR, GDDR2, GDDR3, GDDR4, GDDR5, GDDR6, GDDR7 and what is called and is listed model by increasing the new upward-compatible kenel model of different from access mode generation according to access speed; Or dynamically Rambus internal memory enumerate form for: RDRAM, XDRDRAM, XDR2DRAM and what is called list model by increasing the new upward-compatible kenel model of different from access mode generation according to access speed; Or static RAM (StaticRandom-AccessMemory, SRAM) or only rdma read (Read-OnlyMemory, ROM) or flash memory (FlashMemory) enumerate form for: NORFlash, NANDFlash, 3DNANDFlash, V-Flash, SLC, MLC, eMMC, TLC and what is called list model by increasing the new upward-compatible kenel model of different from access mode generation according to access speed.

Moreover, hard disk memory device (12) can use the hard disk, solid state hard disc, networking hard disk, SAS interface hard disk, SATA interface hard disk, mSATA interface hard disk, mPCIE interface hard disk, FC interface hard disk, scsi interface hard disk, ata interface hard disk, NANDFlash add-on card, the FCoE interface hard disk that comprise traditional magnetic head and disk, and what is called lists model by increasing the new upward-compatible kenel hard disk of different from access mode generation according to access speed.

Again wherein, networking card can select Etherloop, Fast Ethernet road, gigabit Ethernet road, optical fiber, token-ring network, InfiniBand, FCoE (FibreChanneloverEthernet) or Wi-Fi, and its networking speed can adopt 2Mbit/s according to the difference of Internet protocol, 10Mbit/s, 11Mbit/s, 40Mbit/s, 54Mbit/s, 80Mbit/s, 100Mbit/s, 150Mbit/s, 300Mbit/s, 433Mbit/s, 1000Mbit/s, 1Gbit/s, 8Gbit/s, 10Gbit/s, 16Gbit/s, 32Gbit/s, 40Gbit/s, 56Gbit/s, 100Gbit/s, 160Gbit/s, the networking speed of more than 1000Gbit/s and the networking card of any novel networking communications protocol.

Wherein, mainboard can any framework of compatible x86, x86-64, IA-64, Alpha, ARM, SPARC32and64, PowerPC, MIPS, Tilera any one processor compatibility and brand of computer manufacturer produce comprise BeagleBoneBlack, RaspberryPi framework mainboard.

It should be noted that operating system file form in addition, can be AdvFS, Be file system (BFS), Btrfs, Coda, CrossDOS, CD deposit system (DFS), Episode, EFS, exFAT, ext, FAT, global file system (GFS), hierarchical file system (HFS), HFSPlus, HPFS (High-Performance File System), the general parallel file system of IBM, JFS, Macintosh File System, MINIX, NetWare file system, NILFS, Nove11 store-service, NTFS, QFS, QNX4FS, ReiserFS (Reiser4), SpadFS, UBIFS, Unix file system, Veritas file system (VxFS), VFAT, optional position write archives layout (WAFL), XFS, Xsan, ZFS, CHFS, FFS2, F2FS, JFFS, JFFS2, LogFS, NVFS, YAFFS, UBIFS, DCE/DFS, MFS, CXFS, GFS2, Google file system, OCFS, OCFS2, QFS, Xsan, AFS, OpenAFS, AFP, MS-DFS, GPFS, Lustre, NCP, NFS, POHMELFS, Hadoop, HAMMER, SMB (CIFS), cramfs, FUSE, SquashFS, UMSDOS, UnionFS, configfs, devfs, procfs, specfs, sysfs, tmpfs, WinFS, EncFS, EFS, ZFS, RAW, ASM, LVM, SFS, MPFS, MGFS one of them.

And the Internet protocol transmission of entity, to select Etherloop, Fast Ethernet road, gigabit Ethernet road, optical fiber, token-ring network, SS7, GSM, GPRS, EDGE, HSPA, HSPA+, CDMA, WCDMA, TD-WCDMA, LTE, GSM, cdmaOne, CDMA2000, UMTSWCDMA, TD-SCDMA, WiMAX, 3G report net, CDMA20001X, Wi-Fi, SuperwiFi, wi-FiGO and upward-compatible IEEE make new network transmission agreement.

Under system architecture method proposed by the invention, we adopt distributed memory stocking system, the expansion of four kinds of demands can be met, i.e. the expansion that the expansion of the expansion of networking bandwidth for transmission, disk size, the expansion of IOPS speed, internal memory I/O transmission speed increase.Can accomplish to meet the use of trans-regional, data center, WAN simultaneously, user's request can be concentrated at the memory disk of trooping (MemoryDiskCluster) of locality service is provided, also can gradual expansion to troop memory disk capacity, trans-regional or transnational Information Service is provided further.

In the increase along with reservoir, often increase a station server just networking bandwidth of a cumulative machine and disk size, as formation one resource pool (ResourcePool), this dispersion memory type storage facilities unit (1A) is just as an entity hard disk, wherein overall operation can not be affected during an entity host fault, in duplicate, block memory disk (11) understands again backup data to new block memory disk (11), maintain certain data backup basis, also take continuous data to copy (ContinuousDataProtector simultaneously, CDP) for the backup of enterprise's data and reply provide innovative technology, to abolish tape backup often failure, a problem within one day, can only be backed up.

Also allow duplicate generation data can be sent by different blocks memory disk (11) and reach many-one data transmission, when number of users increases, as long as increase block memory disk (11) quantity, the transmission of multi-to-multi can be reached, most hard-disk crashes of RAID are allowed to cause the shortcoming of overall data loss, the crowded delay caused of data that the restriction of reservoir network interface quantity and speed causes user too much, LUN expands the problem of quantity, data center cannot the problem of transregional use, because adopt memory type disk system, each archives or each virtual machine can be present in internal memory by archives form, the I/O high speed that direct use internal memory bus-bar is the highest, data is moved mobile in CPU and internal memory, I/O number, distance, speed is all the highest, take a broad view of above-mentioned advantage, be for known techniques institute can not and person, therefore can its incremental place of See.

The embodiment of the above is only technological thought and feature that the present invention is described, the content that its object enables the person skilled in the art understand the present invention is also implemented according to this, when can not with the scope of the claims of restriction the present invention, namely the equalization generally done according to disclosed spirit changes or modifies, and must be encompassed in the scope of the claims of the present invention.

Take a broad view of above-mentioned, can See the present invention break through under previous technology, really institute's effect for enhancement has been reached, and also non-be familiar with this those skilled in the art institute easily full of beard and, the progressive of its tool, practicality, the aobvious application important document having met patent, so propose patented claim in accordance with the law, earnestly ask your office and check and approve this part application for a patent for invention case, to encourage invention, to feeling moral just.

Claims (28)

1. a distributed memory disk cluster stocking system How It Works, is characterized in that:

First do building of dispersion memory type storage facilities, it is comprise the computer unit of plural platform composition cluster architecture and form memory disk of trooping; Then in computer unit installation system virtual machine platform operating system, installing by this makes computer unit form multiple virtual machine, and the mode that the setting of this computer unit utilizes operating system or program software to carry out set memory capacity to take, this internal memory is formulated for storage device, makes plurality of blocks memory disk; And then each archives is splitted into the data that arrives many parts, keeping the duplicate of to many parts to copy average mark is dispersed on all block memory disks, and use multi channel internal memory bus-bar parallel access memory modules, and become disk to use the capacity planning of memory modules, and wherein all operations system file form is supported in the access of this memory modules, and adopt and distributedly deposit framework, can be more than duplicate copies portion by data; When the operating system of virtual machine directly accesses by CPU the archives needed in memory modules, deposit back after process data in memory modules, virtual machine does the internal memory of calculating use also in memory modules, calculated the memory modules position of still directly depositing go back in deposited VME operating system, to allow in this stocking system data can in computer unit in the mode of parallel calculation in block memory disk process data; Other not the data system of computer unit through networking link connect a Port group now with chain to access the block memory disk of other computer units; And each computer unit is further according to functional one first data center, at least one second data center and the backup center divided into; Wherein feel concerned about with a virtual cluster data master station keyholed back plate in this first data backup, wherein respectively this second data center is with a virtual cluster data backup station keyholed back plate, and wherein this backup center system is with virtual cluster data backup station keyholed back plate, and wherein this first data center and this second data center form a distributed memory file system jointly; In addition, further for forming stack construction expansion storage volume framework, it uses the access mode of networking layer interface block memory disk multiple in a computer unit to be formulated for the memory disk element resources pond of trooping of a concept of trooping, and all block memory disks of synchronous use obtain data; After cluster architecture produces, each cluster architecture can both independently use, and each cluster architecture can be used as dispersion memory type storage facilities unit, use networking layer interface storehouse simultaneously, each dispersion memory type storage facilities unit can be emulated into large-scale memory disk, and the treatment capacity adopting new troop data master station and data backup station of trooping to carry out control data is dispersed in all large-scale memory disks.

2. distributed memory disk cluster stocking system How It Works as claimed in claim 1, it is characterized in that: respectively this block memory disk system is electrically connected at least one hard disk memory device respectively, this hard disk memory device system can the data an interval time by this block memory disk backup.

3. distributed memory disk cluster stocking system How It Works as claimed in claim 1; it is characterized in that: the block memory disk in all computer units; capital uses continuous data protected mode, is unremittingly backed up in unified large-scale physical hard disk cluster disk array by data backup continuously.

4. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: this computer unit system has had the calculator device of CPU, internal memory, hard disk, networking card, mainboard, I/O adapter, connecting line and shell.

5. distributed memory disk cluster stocking system How It Works as claimed in claim 1, it is characterized in that: every part of duplicate can both adopt 1-4096 position AES and RSA to use encrypted data with to leave on internal memory, whenever will accessing data, data flows between internal memory and CPU, virtual machine system is present in memory modules with archives form, virtual memory plan that the memory size of use is also at same block.

6. distributed memory disk cluster stocking system How It Works as claimed in claim 1, it is characterized in that: respectively this block memory disk through a monitoring unit to do the monitoring of its function situation, this detecting unit utilizes the software program of Splunk or other search engines to monitor it, more can configure when a problem is detected and restart application program service, reach the function of a reply.

7. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: virtual machine platform operating system adopts version after VMwarevSphereESXi4.1, the version after MicrosoftServer2012R2Hyper-V, CitrixXenServerOracleVM, OeacleVM, RedHatKVM or Linux-BaseKVM.

8. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: VME operating system adopts Linux, FreeBSD, OpenSolaris or MacOSX.

9. distributed memory disk cluster stocking system How It Works as claimed in claim 1, it is characterized in that: be use storage area networking framework to run between the internal memory of virtual machine, take software virtual networking layer interface out, link together all block memory disks Collaboration.

10. distributed memory disk cluster stocking system How It Works as claimed in claim 9, it is characterized in that: this networking layer interface, system's use SAN, SANiSCSI, SANFC, SANFCoE, NFS, NAS, JBOD, CIFS or FUSE interface carries out the communication between server and disk drive equipment, and exploitation has RAMSTORAGEAPI to use as application program support.

11. distributed memory disk cluster stocking system How It Works as claimed in claim 10, it is characterized in that: this RAMSTORAGEAPI system adopts REST, Restful, C++, PHP, Python, Java, the RAMSTORAGEAPI that Perl or Javascript program development software (PDS) is formed, access dispersion memory type storage facilities unit AP I function, it comprises fault-tolerant, backup, transfer, quick arrangement virtual machine, hyperdisk size, automatically block memory disk is automatically increased according to actual demand capacity, data load balance transfer between block, the function that backup is replied, continuous data is protected, snapshot and monitoring resource.

12. distributed memory disk cluster stocking system How It Works as claimed in claim 1, it is characterized in that: the resource of untapped CPU, internal memory, actual physics hard disk, system uses virtual machine platform operating system that all MRP are become unified resource pool, make computer resource required for each, can adjust and be passed to the more abundant computer unit of other resources.

13. distributed memory disk cluster stocking system How It Works as claimed in claim 1, it is characterized in that: the connected mode of multiple dispersion memory type storage facilities unit is according to the Internet protocol transmission of entity, and adopt SSLVPN, VPN or encryption calculation mode to transmit package, when networking line is obstructed, each region can both operate separately, when line is replied, data by complete synchronization to each block memory disk in each dispersion memory type storage facilities unit.

14. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: this CPU is x86, x86-64, IA-64, Alpha, ARM, SPARC32and64, PowerPC, MIPS, Tilera a kind of processor wherein.

15. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: the internal memory use-pattern of this computer unit be directly by CPU Memory Controller Hub directly or hyperchannel many with threeway with speed direct access internal document more than 800MHz to 1333MHz.

16. distributed memory disk cluster stocking system How It Works as claimed in claim 15, is characterized in that: memory size is 1MB to 16ZB.

17. distributed memory disk cluster stocking system How It Works as claimed in claim 15, is characterized in that: the internal memory kind adopted is dynamic asynchronous memory dram, dynamic synchronization memory dram, dynamic mobile platform internal memory, DYNAMIC DRAWING process internal memory, dynamically Rambus internal memory, static RAM, only rdma read or flash memory.

18. distributed memory disk cluster stocking system How It Works as claimed in claim 17, is characterized in that: dynamic asynchronous memory dram is FPMRAM or EDORAM.

19. distributed memory disk cluster stocking system How It Works as claimed in claim 17, is characterized in that: dynamic synchronization memory dram is SDRAM, DDRSDRAM, DDR2SDRAM, DDR3SDRAM, DDR4SDRAM or DDR5SDRAM.

20. distributed memory disk cluster stocking system How It Works as claimed in claim 17, is characterized in that: dynamic mobile platform internal memory is LPDDR, LPDDR2, LPDDR3 or LPDDR4.

21. distributed memory disk cluster stocking system How It Works as claimed in claim 17, is characterized in that: save as VRAM, WRAM, MDRAM, SGRAM, SDRAM, GDDR, GDDR2, GDDR3, GDDR4, GDDR5, GDDR6 or GDDR7 in DYNAMIC DRAWING process.

22. distributed memory disk cluster stocking system How It Works as claimed in claim 17, is characterized in that: dynamically save as RDRAM, XDRDRAM or XDR2DRAM in Rambus.

23. distributed memory disk cluster stocking system How It Works as claimed in claim 17, wherein flash memory is NORFlash, NANDFlash, 3DNANDFlash, V-Flash, SLC, MLC, eMMC or TLC.

24. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: hard disk memory device is traditional magnetic head, the hard disk of disk, solid state hard disc, networking hard disk, SAS interface hard disk, SATA interface hard disk, mSATA interface hard disk, mPCIE interface hard disk, FC interface hard disk, scsi interface hard disk, ata interface hard disk, NANDFlash add-on card or FCoE interface hard disk.

25. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: networking card is Etherloop, Fast Ethernet road, gigabit Ethernet road, optical fiber, token-ring network, InfiniBand, FCoE or Wi-Fi.

26. distributed memory disk cluster stocking system How It Works as claimed in claim 25, is characterized in that: networking speed adopts the networking speed of 2Mbit/s, 10Mbit/s, 11Mbit/s, 40Mbit/s, 54Mbit/s, 80Mbit/s, 100Mbit/s, 150Mbit/s, 300Mbit/s, 433Mbit/s, 1000Mbit/s, 1Gbit/s, 8Gbit/s, 10Gbit/s, 16Gbit/s, 32Gbit/s, 40Gbit/s, 56Gbit/s, 100Gbit/s, 160Gbit/s or more than 1000Gbit/s.

27. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: any one processor in mainboard compatible x86, x86-64, IA-64, Alpha, ARM, SPARC32and64, PowerPC, MIPS, Tilera.

28. distributed memory disk cluster stocking system How It Works as claimed in claim 1, is characterized in that: operating system file form is AdvFS, BFS, Btrfs, Coda, CrossDOS, DFS, Episode, EFS, exFAT, ext, FAT, GFS, HFS, HFSPlus, HPFS (High-Performance File System), the general parallel file system of IBM, JFS, Macintosh File System, MINIX, NetWare file system, NILFS, Nove11 store-service, NTFS, QFS, QNX4FS, ReiserFS, SpadFS, UBIFS, Unix file system, VeritasVxFS, VFAT, WAFL, XFS, Xsan, ZFS, CHFS, FFS2, F2FS, JFFS, JFFS2, LogFS, NVFS, YAFFS, UBIFS, DCE/DFS, MFS, CXFS, GFS2, Google file system, OCFS, OCFS2, QFS, Xsan, AFS, OpenAFS, AFP, MS-DFS, GPFS, Lustre, NCP, NFS, POHMELFS, Hadoop, HAMMER, SMB (CIFS), cramfs, FUSE, SquashFS, UMSDOS, UnionFS, configfs, devfs, procfs, specfs, sysfs, tmpfs, WinFS, EncFS, EFS, ZFS, RAW, ASM, LVM, SFS, MPFS or MGFS.