CN111722797A

CN111722797A - SSD and HA-SMR hybrid storage system oriented data management method, storage medium and device

Info

Publication number: CN111722797A
Application number: CN202010420508.2A
Authority: CN
Inventors: 伍卫国; 张驰; 张晨; 聂世强; 郑旭达; 马春苗
Original assignee: Xian Jiaotong University
Current assignee: Xian Jiaotong University
Priority date: 2020-05-18
Filing date: 2020-05-18
Publication date: 2020-09-29
Anticipated expiration: 2040-05-18
Also published as: CN111722797B

Abstract

The invention discloses a data management method, a storage medium and equipment for an SSD and HA-SMR hybrid storage system, which orient IO streams to different storage levels by taking the principle that whether an LBA (logical block addressing) of a target data block is aligned with a write pointer of a zone pair in the SMR and combining the size of a non-sequential write request. And for the management strategy of the cache layer, dividing the data blocks in the cache layer into different set sets according to whether the data blocks belong to the same zone, calculating the comprehensive heat of write and read flag bits set for the data blocks entering the cache when cache cleaning operation occurs, calculating the coverage rate of the sets corresponding to different zones, obtaining the comprehensive weight of the set sets by combining the write and read flag bits, and adding the comprehensive weight into the corresponding cache elimination linked list. The invention fully considers the factors of the SSD service life, the cache hit rate and the SMR write amplification, reduces the response delay to the upper application request and improves the overall performance of the hybrid storage system.

Description

SSD and HA-SMR hybrid storage system oriented data management method, storage medium and device

Technical Field

The invention belongs to the technical field of computer storage, and particularly relates to a write-friendly data management method, a storage medium and write-friendly data management equipment for an SSD and HA-SMR hybrid storage system.

Background

As a new type of inexpensive magnetic medium suitable for mass storage of cold data in recent years, SMR shingled disks have the advantages of low price, high storage density, and simple production process, and are suitable for mass data storage today, but as the capacity of the disks increases, the disks are overlapped by compressing magnetic tracks, and a larger and wider magnetic head is required to generate a stronger magnetic field. SMR is no different from a general HDD disk in terms of read requests and sequential write requests, but at the time of random writing, since a magnetic head interferes with a track adjacent to a target track downstream to cause erasure of downstream data, a read-modify-write operation needs to be performed on the data each time, and a so-called "write amplification" caused thereby causes a response delay to IO applied to an upper layer. The SSD is generally composed of flash chips, has stronger random read-write capability compared with a magnetic disk, and can effectively cope with random IO of upper-layer services by using the SSD as a cache.

However, SSDs also have asymmetric read and write characteristics, and due to the limited number of erase operations, frequent erase operations can cause "write through" of the flash memory chip, which can cause irreversible effects on the stability of data storage. Meanwhile, compared with the early SMR, the SMR with the host management mode and the host sensing mode can expose the internal state and the device information of the disk to the upper layer application so as to help the upper layer application to better know the bottom layer information. However, the data management policy of the existing hybrid storage system is usually only to simply use the LRU algorithm to eliminate data that is not used for a long time or write data into the SSD as much as possible based on the locality principle, so that the cache layer has a high hit rate, which does not fully consider the characteristics of the underlying storage medium, and has a great influence on the service life and the overall performance of the system.

Therefore, how to enable the upper layer application to better organize and sequence the service IO stream by using the internal information of the HA-SMR (host-aware shingled disk) and reduce the write-in of the invalid flow of the SSD so as to prolong the service life of the SSD; and how to use the information to make the SSD serving as the cache layer formulate a proper cache replacement strategy to reasonably write back data with different cold and hot states to the SMR, so that the influence of SMR write amplification on upper-layer business IO is reduced, and the problem of efficient data management of the hybrid storage system needs to be considered.

Disclosure of Invention

The technical problem to be solved by the present invention is to provide a write-friendly data management method for a mixed storage system of SSD and HA-SMR, which is a mixed storage system composed of SSD as a cache layer and HA-SMR as a capacity layer, and which solves the application service scenario of a read request of a mixed part mainly based on a write request, and completes the implementation at a host end by considering factors such as SSD life, cache hit rate, SMR write amplification; when data enters the SSD, judging whether a write path is directly written into a capacity layer or a cache layer according to write pointer state information of the HA-SMR; meanwhile, when the data of the cache layer is written back to the capacity layer, the data belonging to the same zone is written back together, on one hand, the available space of the cache layer is improved, on the other hand, the cleaning time of a lasting buffer layer in the SMR is reduced, more random writing is converted into sequential writing, and the SMR writing amplification rate is reduced. Thereby improving the performance of the hybrid storage system as a whole.

The invention adopts the following technical scheme:

a write-friendly data management method facing an SSD and HA-SMR hybrid storage system comprises the following steps:

s1, clustering blocks belonging to the same zone in the SMR into a set in the SSD cache layer, and inserting the blocks into the set in an LBA ascending order mode_blockManaging a linked list, and writing back to an SMR capacity layer by taking set as a basic unit; inserting set into zone in FIFO first-in first-out mode_setA temporary management linked list;

s2, respectively setting a write flag bit and a read flag bit for each block in the set, recording the reading and writing times of the block, setting an initial value to be 0, and adding 1 each time after the reading and writing request hits;

s3, calculating the comprehensive heat of each block in the updated set according to the reading and writing times recorded in the step S2; simultaneously calculating the coverage rate of the updated set;

s4, calculating and updating the final weight of each set according to the comprehensive heat and coverage rate of each block in the step S3, and adding the weight to the zone again in a weight descending manner_setCaching the elimination linked list;

s5, when the remaining space of the buffer is less than the prescribed threshold, the zone from step S4_setSelecting an elimination object from the cache elimination linked list, executing a cache replacement algorithm, replacing data meeting the requirement from the cache to a capacity layer, and executing the step S6 after the replacement is finished;

s6, clearing the write and read flag bits of the replaced data block in the SSD cache layer;

s7, when the read request arrives, firstly checking whether the requested block exists in the cache, if so, reading the target data from the SSD cache layer and returning the target data to the upper layer application, and then executing the step S2; if the read request is not hit in the cache, reading target data from the SMR capacity layer and returning the target data to the upper application;

s8, when the write request arrives, firstly checking whether the requested data block exists in the cache, if so, directly updating the target data in the SSD cache layer, and then executing the step S2; if the target data does not exist in the cache layer, judging whether the LBA of the target data block is aligned with a write pointer of a zone pair in the SMR, and if the LBA of the target data block is aligned with the write pointer of the zone pair in the SMR, directly writing the zone pair in the SMR; otherwise, judging whether the size of the non-sequential write request exceeds the set non-sequential write threshold value T or not_NSWIf the number of the SMR data exceeds the preset value, the SMR data is regarded as large non-sequential writing, and the large non-sequential writing is written into a persistent buffer area inside the SMR disk; if the two cases are not met, the data needs to be written into the SSD cache layer, and steps S1 and S2 are executed; judging whether the residual space is less than the residual space threshold T or not after the write request is completed_freeIf it is less than the threshold value T_freeThen step S5 is executed to implement write-friendly data management of the SSD and HA-SMR hybrid storage system.

Specifically, in step S1, the minimum unit of the upper IO request is a 4KB block, and the HA-SMR uses a 256 MB-sized zone as a management unit and consists of 65536 blocks with a size of 4 KB; the set is composed of all block blocks belonging to the same zone and located in the SSD, and each block in the set is addressed according to the LBAAscending insertion set_blockManaging a linked list; each set in the SSD corresponds to one zone in the SMR, the set is a minimum of 4KB and a maximum of 256MB, and is inserted into the zone in FIFO form for each set_setThe linked list is temporarily managed.

Specifically, in step S3, a write flag bit and a read flag bit are added to each block entering the cache, the update and read times are recorded, and the comprehensive heat of each block in the cache is calculated.

Specifically, in step S3, the coverage of the updated set is C_i＝set_block/zone_block，0≤C _i1 or less, zone _ block is constant, representing 65536 blocks of 4KB size in 256MB size zones; set _ block represents the number of 4KB sized blocks in the set, 0 ≦ set _ block ≦ 65536.

Specifically, in step S4, the total heat of the block and the coverage of the set are obtained in step S3, the total weight of the sets is obtained, and the zone is inserted into each set in descending order of the total weight_setManaging a linked list with each node including an integrated weight W_iAnd coverage rate C_iThe information of (1).

Specifically, step S5 specifically includes:

s501, when the read request is reached, checking whether a block of the request exists in a cache or not according to a mapping table, if so, reading target data from an SSD cache layer and returning the target data to an upper layer application, and then executing the step S502;

s502, updating the read zone bit of the block which hits the reading request, and adding 1 to the recording times;

and S503, when the read request is not hit in the cache, reading the target data from the SMR capacity layer and returning the target data to the upper application.

Specifically, step S6 specifically includes:

s601, when a write request arrives, checking whether a requested data block exists in a cache according to a mapping table, if so, directly updating target data in an SSD cache layer, and then executing a step S602;

s602, updating the write zone bit of the block which hits the write request, and adding 1 to the recording times;

s603, if the target data does not exist in the cache layer, judging whether the LBA of the target data block is aligned with the write pointer of the zone pair in the SMR, and if so, directly writing the target data block into the zone in the SMR, otherwise, entering the step S604;

s604, according to the judgment result of S603, if the LBA of the target data block is not aligned with the write pointer of the zone pair in the SMR, judging again whether the magnitude of the non-sequential write request exceeds the set non-sequential write threshold TNSW, and if the magnitude exceeds T_NSWIf the write command is 1MB, the write command is regarded as large non-sequential write, and the write command is written into a persistent buffer inside the SMR disk, otherwise, the step S605 is executed;

s605, writing data into the SSD cache layer because the conditions of the step S603 and the step S604 are not satisfied, and entering the step S606;

s606, adding a write zone bit and a read zone bit to a block newly entering an SSD cache layer, and adding 1 to the write zone bit recording times; subsequently, if the corresponding set already exists_blockThe management linked list is added into the linked list, otherwise a new set is created_blockManaging linked list, adding new set into zone after completing the above steps_setA temporary management linked list;

s607, after the write request is finished, judging whether the residual space is less than the residual space threshold value T_freeI.e. T _free10% of SSD buffer capacity, if less than threshold T_freeThen, step S7 is executed to start the cache scrubbing operation.

Specifically, step S7 specifically includes:

s701, calculating the comprehensive heat of the block blocks in each set according to the updating times recorded in the step S2, and calculating the coverage rate of each set; subsequently, the zone is calculated_setTemporarily managing the final comprehensive weight of each set in the linked list;

s702, adding the set sets to the zone in sequence according to the rule that the comprehensive weight value is descending from large to small_setEliminating the linked list;

s703 slave zone_setAt the end of the culling list, i.e. beginning at the end with the lowest weight, the set of each node is checked first_blockWhether a write pointer pair of a starting LBA and a zone to be written back is existed in the management chain tableIf the candidates having the coverage rate Ci of 10% are present, they are written back to the zone of SMR as eliminators in a sequential writing manner until the capacity of the cache space returns to the remaining space threshold T_freeAbove; if no such candidate exists, proceed to step S704;

s704, again starting from the zone_setStarting from the end of the linked list, checking the coverage rate information of each node, writing a set with the coverage rate Ci reaching more than 50% as a candidate back to a persistent buffer in the SMR until the capacity of the cache space returns to the residual space threshold T_freeAbove; at this time, the write-back operation is changed into non-sequential write, the device side executes read-modify-write operation to sequentially write the data of the persistent buffer back to the zone of the SMR, and if the condition is not met, the step S705 is performed;

s705, if no candidate exists, the slave zone_setStarting from the end of the linked list, selecting the node with the lowest weight as a candidate to write back to a persistent buffer in the SMR until the capacity of the cache space returns to the residual space threshold value T_freeAbove; at this time, the write-back operation is changed into non-sequential write, and the device side executes read-modify-write operation to write the data in the persistent buffer back to the zone of the SMR in sequence.

A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of claims 1-8.

A management device, comprising:

one or more processors, memory, and one or more programs stored in the memory and configured for execution by the one or more processors, the one or more programs including instructions for performing any of the methods of claims 1-8.

Compared with the prior art, the invention has at least the following beneficial effects:

the invention aims at mixed storage of SSD and HA-SMRThe write-friendly data management method of the storage system is based on the fact that the service life of the limited erasing times of the SSD and the fact that the SMR random writing can generate ' write amplification ', on one hand, the SSD is used as a cache layer due to the fact that the SSD has stronger random writing capability, and on the other hand, the sequential ' IO requests meeting the writing pointer alignment condition are directly directed to the SMR on the basis of the principle that whether the IO request initial address of the upper layer is aligned with the writing pointer of the SMR or not, so that the sequential writing capability of the mechanical hard disk is better utilized; for write pointer misalignment but exceeding a set threshold T_NSWThe non-sequential writing can be written into a persistent buffer area in the SMR disk, the write flow with low SSD value is further filtered, the writing frequency of the SSD is reduced through the steps, frequent erasing is avoided, and the service life of the SSD is prolonged; and finally, only small non-sequential writes can enter the SSD buffer layer with good random write capability, and blocks of the same zone are clustered and reordered to be converted into larger continuous blocks which are beneficial to the execution of 'read-modify-write' cleaning operation of the persistent buffer layer in the HA-SMR, so that the PRT-system performance recovery time is shortened, and the write amplification rate of the SMR is reduced.

Furthermore, by clustering blocks belonging to the same zone and writing the blocks back to the SMR together, the frequency of the read-modify-write operation executed by the equipment end can be reduced, the cleaning efficiency of a persistent buffer area of the equipment end is improved, the PRT performance recovery time of a disk is shortened, the write amplification rate of the SMR is reduced, and the overall performance of the system is improved.

Furthermore, by adding read and write flag bits to the block, the read-write characteristics of the block can be distinguished with finer granularity.

Furthermore, the reading heat and the writing heat of the block can be respectively calculated through the recorded reading and writing times, so that the comprehensive heat of the block can be solved; and calculating the number of blocks in the cache by using the set corresponding to a certain zone to obtain the corresponding coverage rate, thereby obtaining the occupation ratio of the certain zone in the cache.

Further, by calculating the comprehensive heat of each block in the set and the coverage of the set corresponding to the zone in the cache, the heat and quantity factors of the data blocks are comprehensively considered, and the comprehensive weight of the set is obtained according to the heat and quantity factorsAdding the set into the zone in a descending manner according to the weight value_setThe obsolete linked list is cached.

Furthermore, when cache replacement occurs, the overall weight value obtained by calculation is used for obtaining the zone_setAnd selecting the best culler from the tail end of the cache culling linked list.

Further, the data block replaced out of the SSD cache layer will not enter the cache layer again from the capacity layer, and therefore the write and read flag bits thereof are cleared.

Further, when the read request is reached, the data storage position is located by searching the mapping table, and the data is read from the corresponding layer and returned to the upper application.

Furthermore, when a write request arrives, a mapping table is searched for a specific storage position of the data, if cache hit, direct write is performed, otherwise, whether the zone of the SMR is written is determined according to whether the LBA logical address of the data is aligned with the write pointer, and for larger random write, the data is written into the persistent buffer area of the SMR, and only small and inevitable random write is written into the SSD cache layer.

In summary, the invention comprehensively considers three factors, namely the cache hit rate, the SSD service life and the SMR write amplification, so as to improve the comprehensive performance of the hybrid storage system as much as possible.

The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.

Drawings

FIG. 1 is a schematic diagram of a hybrid storage system architecture of the present invention;

FIG. 2 is a data flow diagram of the present invention

FIG. 3 is a diagram of a FIFO managed zone of the present invention_setA temporary management linked list schematic diagram;

FIG. 4 is a block diagram of the LBA descending tube according to the present inventionHair care zone_setA schematic diagram of a cache elimination linked list;

FIG. 5 is a flow chart of the present invention process for handling a read I/O request;

FIG. 6 is a flow chart of the present invention process for handling a write I/O request;

FIG. 7 is a flow chart of cache replacement operation according to the present invention.

Detailed Description

Referring to fig. 1, the management method of the present invention is implemented in a single kernel module in the overall logic structure of the hybrid storage system, so that the Linux operating system does not need to be modified. The module is positioned in the block layer in a form of pseudo block equipment, mainly comprises cache management and IO redirection functions, and is respectively used for triggering data write-back operation during cache space cleaning and responding to read-write requests under different conditions to realize IO relocation.

Referring to fig. 2, for the specific data flow of the present invention, when a read/write request arrives, the storage locations of different types of data flow are different, or when cache replacement occurs, the data flow is also changed.

The invention provides a write-friendly data management method facing an SSD and HA-SMR hybrid storage system, which positions different types of requests to different storage layers by judging data flow direction and reduces the data meeting the requirements to be written back to a capacity layer when cache cleaning operation is triggered. The method comprises the following specific steps:

s1, clustering blocks belonging to the same zone in the SMR into a set in the SSD cache layer, and inserting the blocks into the set in an LBA ascending order mode_blockManaging the linked list and writing back to the SMR capacity layer in a basic unit of set. Inserting set into zone in FIFO first-in first-out mode_setTemporary management link lists, as shown in FIG. 3;

the minimum unit of the upper layer IO request is a block of 4KB, and HA-SMRs each have a 256MB size zone as a management unit, and consist of 65536 blocks of 4KB size. That is, a set is composed of all block blocks belonging to the same zone and located in the SSD, and for each block inside the set, the set is inserted in the ascending order of LBA addresses_blockA linked list is managed. Therefore, the temperature of the molten metal is controlled,each set in the SSD corresponds to a zone in the SMR, the set is a minimum of 4KB and a maximum of 256MB, and is inserted into the zone in the form of a FIFO for each set_setThe linked list is temporarily managed.

s3, calculating the comprehensive heat of each block in the updated set according to the reading and writing times recorded in S2; simultaneously calculating the coverage rate of the updated set;

in order to make the cache have higher hit rate for the read-write requests, therefore, the write and read flag bits are added to each block entering the cache, and the update and read times are recorded respectively, using formula H_i＝W_i+X*R_iCalculating the comprehensive heat of each block in the cache, wherein 0<X<1, X represents the proportion of read requests in the scene that the write requests are main mixed part read requests. At the same time, W_iAnd R_iThe initial values are all 0, which represents the number of updates and reads after the block enters the cache.

Further, set coverage indicates the proportion of the block belonging to a certain zone in the SMR in the cache layer, and can be represented by formula C_i＝set_block/zone_blockSpecific data value can be obtained, C is more than or equal to 0_iLess than or equal to 1. Wherein, zone_blockConstant, representing 65536 blocks of size 4KB in 256MB size zones; and set_blockRepresents the number of 4 KB-sized blocks in the set, where 0 ≦ set_block≤65536。

S4, calculating and updating the final weight of each set according to the comprehensive heat and coverage rate of each block in S3, and adding the weight to the zone again in a weight descending manner_setCaching the elimination linked list;

referring to FIG. 4, the integrated heat of the block and the coverage of the set obtained in step S3 are calculated according to the formula W_i＝∑H_i/C_iDetermining a comprehensive weight of set, wherein C_iDenotes the ith set_iAnd ∑ H_iPresentation cachingSet i of the ith set_iThe sum of the comprehensive heat of all the block blocks; further, each set is inserted into the zone according to the descending order of the magnitude of the comprehensive weight_setManaging a linked list with each node including an integrated weight W_iAnd coverage rate C_iThe information of (1).

S5, when the remaining space of the buffer is less than the prescribed threshold, the zone from S4_setSelecting an elimination object from the cache elimination linked list, executing a cache replacement algorithm, replacing data meeting the requirement from the cache to a capacity layer, and executing the step S6 after the replacement is finished;

referring to fig. 5, for analyzing the data flow of the IO redirection function, if a read request arrives, the processing of the read request includes the following steps:

s501, when a read request is reached, firstly checking whether a requested block exists in a cache according to a mapping table, if so, reading target data from an SSD cache layer and returning the target data to an upper layer application, wherein the data flow path of the target data is as a black dotted line in the graph 2, and then executing the step S502;

s503, when the read request is not hit in the cache, reading target data from the SMR capacity layer and returning the target data to an upper layer application, wherein the data may exist in a persistent buffer layer or a zone of the SMR, the data flow path of the data flow path returns to the upper layer along the reference numeral 2 or the reference numeral 3 in the figure 2, and the position of the set in the corresponding linked list does not need to be adjusted.

referring to fig. 6, if a write request arrives, the specific steps are as follows:

s601, when a write request arrives, firstly checking whether a requested data block exists in a cache according to a mapping table, if so, directly updating target data in an SSD cache layer, wherein the data flow direction is as indicated by a reference numeral 1 in FIG. 2, and then executing a step S602;

s603, if the target data does not exist in the cache layer, judging whether the LBA of the target data block is aligned with the write pointer of the zone pair in the SMR, if the conditions are met, directly writing the zone in the SMR, wherein the data flow path is indicated by the reference number 2 in FIG. 2, otherwise, entering the step S604;

s604, according to the judgment result of S603, if the LBA of the target data block is not aligned with the write pointer of the zone pair in the SMR, judging again whether the size of the non-sequential write request exceeds the set non-sequential write threshold T_NSWIf it exceeds T_NSWThe method is regarded as larger non-sequential writing, which is beneficial to improving the cleaning efficiency of the disk persistent buffer, the method can be written into the persistent buffer inside the SMR disk, the data flow path of the persistent buffer is indicated by a mark 3 in figure 2, otherwise, the method enters the step S605;

s605, because the conditions of S603 and S604 are not satisfied, at this time, the data needs to be written into the SSD cache layer, the data flow path is changed to the direction indicated by reference numeral 1 in the figure again, and the process proceeds to step S606;

s606, adding a write flag bit and a read flag bit to the block newly entering the SSD cache layer, and adding 1 to the write flag bit recording times. Subsequently, if the corresponding set already exists_blockAdding the management linked list into the linked list, otherwise, creating new set_blockManaging linked list, adding new set into zone after completing the above steps_setA temporary management linked list;

s607, after the write request is finished, judging whether the residual space is less than the residual space threshold value T_freeIf it is less than the threshold value T_freeThen, step S7 is executed to start the cache scrubbing operation.

After a certain time has elapsed, a persistent buffer within the SMR capacity layer or a clean-up operation of read-modify-write is performed, at which time data is written back to the zone with the data flow indicated by reference number 4 in FIG. 2.

referring to fig. 7, for analyzing the data flow direction of the cache management function, when the remaining space is insufficient and the cache replacement algorithm needs to be executed, the specific steps are as follows;

s701, according to the updating times recorded in the step S2, utilizing a formula H_i＝W_i+X*R_iCalculating the comprehensive heat of block blocks in each set and obtaining the heat through a formula C_i＝set_block/zone_blockCalculating the coverage rate of each set; then, using the formula W_i＝∑H_i/C_iCalculating the zone shown in FIG. 3_setTemporarily managing the final comprehensive weight of each set in the linked list;

s703 slave zone_setAt the end of the culling list, i.e. beginning at the end with the lowest weight, the set of each node is checked first_blockWhether the starting LBA is aligned with the write pointer of the destination zone to be written back or not in the management chain table and the coverage rate C_iThe candidate up to 10%, if any, is written back to the SMR zone as an evictor in a sequential write, the data flow is as shown by reference numeral 5 in FIG. 2, and the scrubbing operation is performed until the cache space capacity returns to the remaining space threshold T_freeAbove. If no such candidate exists, proceed to step S704;

s704, again starting from the zone_setStarting from the end of the linked list, checking the coverage rate information of each node, and comparing the coverage rate C_iThe set up to 50% or more is written back as a candidate to the persistent buffer in SMR, the data flow is shown as number 6 in FIG. 2, and the flush operation is performed until the capacity of the cache space returns to the remaining space threshold T_freeAbove. At this time, the write-back operation will be changed to non-sequential write, when the device side buffer is full or the device is in an idle state, the device side executes read-change-write operation to write the data of the persistent buffer back to the zone of the SMR in sequence, the data flow is as shown by reference numeral 4 in fig. 2, if the above condition is not met, the step S705 is entered;

s705, when the candidates of the two cases do not exist, the slave zone_setStarting from the end of the linked list, the node with the lowest weight is selected as a candidate to be written back to the persistent buffer in the SMR, and the data flow is as shown by reference numeral 6 in FIG. 2 until the capacity of the buffer space returns to the residual space threshold T_freeAbove. At this time, the write-back operation is changed into non-sequential write, and when the device side buffer is full or the device is in an idle state, the device side executes read-write-back operation to sequentially write the data in the persistent buffer back to the zone of the SMR.

S8, when the write request arrives, firstly checking whether the requested data block exists in the cache, if so, directly updating the target data in the SSD cache layer, and then executing the step S2; if the target data does not exist in the cache layer, judging whether the LBA of the target data block is aligned with a write pointer of a zone pair in the SMR, and if the LBA of the target data block is aligned with the write pointer of the zone pair in the SMR, directly writing the zone pair in the SMR; otherwise, judging whether the size of the non-sequential write request exceeds the set non-sequential write threshold value T or not_NSWI.e. T_NSWIf the current value exceeds 1MB, the current value is regarded as larger 'non-sequential writing', and a persistent buffer area inside an SMR disk can be written; if the two cases are not met, at this time, the data needs to be written into the SSD cache layer, and steps S1 and S2 are executed; judging whether the residual space is less than the residual space threshold T or not after the write request is completed_freeI.e. T _free10% of SSD buffer capacity, if less than threshold T_freeThen step S5 is executed.

In summary, according to the write-friendly data management method for the SSD and HA-SMR hybrid storage system, three factors, namely SSD service life, SMR write amplification phenomenon and cache hit rate, are considered, the SSD is used as a cache layer to cache random writing, and the sequential IO requests meeting the write pointer alignment condition are directly directed to the SMR on the principle that whether the IO request start address of the upper layer is aligned with the write pointer of the SMR or not, so that the sequential writing capability of the mechanical hard disk is better utilized; for non-sequential writes with misaligned write pointers but exceeding a set threshold, the writes can be written to a persistent buffer inside the SMR disk, further filtering write traffic that is "of low value" to the SSD, and only writing small, unavoidable non-sequential writes to the SSD buffer. The steps reduce the write-in times of the SSD, avoid frequent erasure, prolong the service life of the SSD, cluster and reorder the blocks of the same zone at the same time, convert the blocks into larger continuous blocks which are beneficial to the execution of 'read-modify-write' cleaning operation of a persistent buffer layer in the HA-SMR, shorten the PRT-system performance recovery time, reduce the write amplification rate of the SMR, and effectively improve the overall performance of the hybrid storage system based on the above modes.

The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.

Claims

1. A data management method for a hybrid storage system of an SSD and an HA-SMR is characterized by comprising the following steps:

s5, when the remaining space of the buffer is less than the prescribed threshold, the zone from step S4_setSelecting obsolete objects from the cache obsolete linked list, executing a cache replacement algorithm, and caching the data meeting the requirementsReplacing to the capacity layer, and executing step S6 after completion;

2. The method according to claim 1, wherein in step S1, the minimum unit of the upper IO request is a block of 4KB, and the HA-SMR comprises 65536 blocks of 4KB with a 256MB size zone as a management unit; the set is composed of all block blocks belonging to the same zone and located in the SSD, and for each block in the set, the set is inserted according to the ascending order of LBA addresses_blockManaging a linked list; each set in the SSD corresponds to one zone in the SMR, the set is a minimum of 4KB and a maximum of 256MB, and is inserted into the zone in FIFO form for each set_setThe linked list is temporarily managed.

3. The method of claim 1, wherein in step S3, a write flag bit and a read flag bit are added to each block entering the cache, the number of updates and reads are recorded, and the comprehensive heat of each block in the cache is calculated.

4. The method according to claim 1, wherein in step S3, the coverage rate of the updated set is C_i＝set_block/zone_block，0≤C_i1 or less, zone _ block is constant, representing 65536 blocks of 4KB size in 256MB size zones; set _ block represents the number of 4KB sized blocks in the set, 0 ≦ set _ block ≦ 65536.

5. The method of claim 1, wherein in step S4, the comprehensive weight of sets is determined by the comprehensive heat of the blocks and the coverage of the set sets obtained in step S3, and each set is inserted into the zone according to the descending order of the comprehensive weight_setManaging a linked list with each node including an integrated weight W_iAnd coverage rate C_iThe information of (1).

6. The method according to claim 1, wherein step S5 is specifically:

7. The method according to claim 1, wherein step S6 is specifically:

s604, according to the judgment result of S603, if the LBA of the target data block is not aligned with the write pointer of the zone pair in the SMR, judging again whether the size of the non-sequential write request exceeds the set non-sequential write threshold T_NSWIf it exceeds T_NSWIf the write command is 1MB, the write command is regarded as large non-sequential write, and the write command is written into a persistent buffer inside the SMR disk, otherwise, the step S605 is executed;

s607, after the write request is finished, judging whether the residual space is less than the residual space threshold value T_freeI.e. T_free10% of SSD buffer capacity, if less than threshold T_freeThen, step S7 is executed to start the cache scrubbing operation.

8. The method according to claim 1, wherein step S7 is specifically:

s703 slave zone_setAt the end of the culling list, i.e. beginning at the end with the lowest weight, the set of each node is checked first_blockWhether a candidate exists in the management linked list, wherein the starting LBA is aligned with the write pointer of the target zone to be written back, and the coverage rate Ci reaches 10%, if the candidate exists, the candidate is used as a deseriant and written back to the zone of the SMR in a sequential writing mode until the capacity of the cache space returns to the residual space threshold value T_freeAbove; if no such candidate exists, proceed to step S704;

9. A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of claims 1-8.

10. A management device, comprising: