CN111857601B - Solid-state disk cache management method based on garbage collection and channel parallelism - Google Patents

Abstract

The invention discloses a solid-state disk cache management method based on garbage collection and channel parallelism, which manages a cache region through a garbage collection detection write mechanism and a channel parallel write mechanism. Before the cache write-back, a garbage collection probe write mechanism is adopted to avoid the I/O conflict as far as possible, and meanwhile, the cache miss cost is reduced. When the cache is written back, a channel parallel writing mechanism is adopted as much as possible, the time cost of write-once operation is hidden by write-once operation, and the writing bandwidth of the solid-state disk is improved. According to the solid-state disk cache management method, by reasonably utilizing the garbage collection detection writing mechanism and the channel parallel writing mechanism, the cache writing time is saved while the conflict between the user I/O and the internal I/O of the solid-state disk is reduced, and the integral writing performance of the solid-state disk is improved.

Description

A Solid State Disk Cache Management Method Based on Garbage Collection and Channel Parallelism

technical field

The invention relates to the technical field of storage systems, in particular to a solid state disk cache management method based on garbage collection and channel parallelism.

Background technique

SSD cache management is an important technical means to improve the storage performance of SSD while adapting to the physical characteristics of flash memory. SSDs internally behave differently from disks, so a redesign of cache management methods within SSDs is necessary. Traditional caching methods mainly include LRU, LFU and their derivative algorithms, which mainly use the time locality of the load to improve the cache hit rate. In addition to improving the cache hit rate, the solid-state disk caching method also needs to be optimized for situations such as inconsistent read and write performance of solid-state disks, unstable write performance, and reduced garbage collection.

The garbage collection detection write mechanism is to redesign the cache write-back strategy for the internal I/O conflicts that may be caused by the garbage collection of the solid-state disk during the cache write-back process. Because the internal I/O caused by garbage collection will make the upper-layer I/O requests have to wait, so when triggering writeback, first detect whether the solid-state disk chip to which the logical page to be written back is mapped is performing garbage collection operations, so as to avoid upper-layer I/O requests. /O conflicts with underlying I/O. Improve cache efficiency by reducing the cost of cache misses, while improving the overall write performance of the SSD storage system.

The channel parallel write mechanism is aimed at the situation that there are independent data and signal buses between the internal channels of the solid state disk, and read and write operations can be performed concurrently. It is proposed to use the parallel write back method between channels to improve the overall write bandwidth of the solid state disk. When triggering the channel parallel writing mechanism, when searching for multiple suitable logical pages for parallel writing back, combined with the information provided by the garbage collection detection writing mechanism, select logical pages that do not exceed the number of SSD channels and belong to different channels for writing back. The write performance of the solid state disk storage system is improved by hiding multiple write operations through one write operation.

In the process of realizing the present invention, the inventor found that there are at least the following technical problems in the prior art: the traditional LRU-based cache optimization algorithm does not fully consider the I/O conflicts that may be generated by garbage collection in the solid state Parallel read and write characteristics between internal channels of SSDs. Therefore, a reasonable consideration of the internal characteristics of the solid state disk can improve the overall storage performance of the solid state disk.

Contents of the invention

The purpose of the present invention is to propose a solid-state disk cache management method based on garbage collection and channel parallelism for the I/O conflicts that may be caused by garbage collection inside the solid-state disk and the rational use of the parallel characteristics between the internal channels of the solid-state disk. The technical solution is mainly Set the temperature attribute and dirty attribute for each logical page that enters the SSD cache, and adjust the values of these two attributes according to the number of hits and mapping. When the cache space is insufficient and the cache writeback is triggered, the logical page for writeback is reasonably selected. When selecting a logical page, first detect whether the physical location of the logical page is being garbage collected according to the garbage collection detection writing mechanism, and then judge whether the logical page is a cold dirty page. When the found logical page satisfies the channel parallel write When the mechanism is used, the searched logical pages are written back at one time. When the channel parallel write mechanism is not satisfied, the corresponding logical pages are written back according to the preset priority order. This method reduces the cost of cache misses and improves the overall write bandwidth.

The purpose of the present invention can be achieved by taking the following technical solutions:

A solid-state disk cache management method based on garbage collection and channel parallelism. This method takes into account the resource conflict between upper-layer application requests and solid-state disk internal garbage collection I/O and the channel-level parallel characteristics of solid-state disks. Two solutions are proposed. Processing mechanism: Garbage collection detection write mechanism and channel parallel write mechanism. The garbage collection detection write mechanism will detect whether the chip to which the logical page is mapped is performing garbage collection operations to avoid I/O conflicts and reduce the cost of cache misses. The channel parallel writing mechanism reasonably writes back the logical pages of the maximum number of channels and different channels at one time, hides multiple write operations through one write operation, and improves the write bandwidth of the SSD.

A solid-state disk cache management method based on garbage collection and channel parallelism. Logical pages in the onboard cache queue of the solid-state disk perform write-back operations according to the preset garbage collection detection write mechanism and channel parallel write mechanism. The I/O to the bottom layer is cut into the physical page size and enters the cache area, which is called a logical page. The described solid state disk cache management method includes the following steps:

S1. After the operating system sends the upper-layer application request from the block device layer to the flash onboard cache, it maintains a temperature attribute for each logical page, which is used to dynamically identify the hot and cold status of the logical page, and at the same time according to the mapping relationship The logical page maintains a clean/dirty attribute, which is used to dynamically identify the clean/dirty state of the logical page;

S2. When the cache space of the solid-state disk is insufficient, the cache write-back operation will be triggered. When traversing each logical page, use the garbage collection detection writing mechanism to detect whether the chip where the physical location of the logical page mapping is located is performing garbage collection operations;

S3. When traversing the cache queue, according to the information provided by the garbage collection detection writing mechanism, the cold/hot status and clean/dirty attributes of the logical page, the number of internal channels N of the solid state disk is taken as the upper limit, and the number of channels not exceeding N is selected. Multiple logical pages, the number is n. When n is greater than 1, trigger channel parallel write operation to write back the selected n logical pages at one time; when n is less than or equal to 1, find a suitable logical page to write back according to the predetermined priority order.

Further, the rules for adjusting the hot and cold states and clean/dirty flags of the logical pages in step S1 are as follows:

When each logical page enters the cache, set an integer temperature attribute for it to identify the temperature state, and then adjust the temperature attribute value of the logical page according to the number of cache hits;

When each logical page enters the cache, a Boolean dirty attribute is set for it to identify the clean/dirty status of the logical page. The logical page initially entered into the cache is a clean page by default, and the dirty attribute status is adjusted according to the mapping relationship.

Further, in the step S2, when the upper-layer request enters the cache area, cache service needs to be performed or cache space needs to be vacated due to insufficient space, and the garbage collection detection writing mechanism will start to work when traversing the cache queue. The main steps are as follows:

S21. When a request arrives, first judge whether it is a hit in the cache area;

S22. When the cache area fails to hit, judge whether it is possible to insert the cache first according to the remaining space of the cache;

S23. When the cache space is insufficient, a cache write-back operation needs to be triggered. When searching for a suitable logical page, a garbage collection detection write mechanism is triggered for each logical page traversed. As shown in Figure 1, the garbage collection detection write mechanism is used It can be seen that the chip where the physical location of the logical page P8 is mapped to is doing garbage collection (P8 page is a GC page), at this time, P8 should be temporarily stored in the cache, and start from P7 to judge whether the logical page is a GC page again , combined with channel parallel write mechanism for logical page search.

Furthermore, during the logical page search process, the logical page search is performed according to the predetermined priority, and the operation steps of the channel parallel writing mechanism are as follows:

S31. When looking for a suitable logical page, start traversing from the end of the cache queue forward, judge whether the logical page is a GC page according to the information provided by the garbage collection detection writing mechanism, and if so, directly traverse the next cache node;

S32. When the logical page traversed is not a GC page, judge whether the logical page is a cold dirty page according to the temperature value and the dirty attribute value of each logical page in the cache, if the logical page is not a cold dirty page, continue to traverse A logical page; if the logical page is a cold and dirty page, continue to judge whether the channel mapped to the logical page is the same as the channel mapped to the logical page in the set of logical pages to be written back, if not, add the logical page In the logical page collection, until the number of logical pages of the channel is found or the entire cache queue is traversed;

S33. When the number of logical page set elements to be written back is greater than 1, trigger the channel parallel write mechanism to write back all the logical pages in the set at one time;

S34. When the number of elements in the set of logical pages to be written back is less than or equal to 1, search for cold pages, hot pages, and GC pages in order from the tail node in the cache queue according to the priority order, and stop searching each time one is found. , and finally write back the found logical pages. As shown in Figure 1, because the logical page P8 is a GC page, it will skip this page directly and start searching from P7. The priority order is: multiple cold and dirty pages>single cold page>single hot page>GC page, logical page P7, P4, and P1 meet the highest priority. This cache emptying operation will write back these three logical pages at one time. Similarly, when the highest priority is not satisfied, the search process of the next priority logical page is triggered at one time.

Further, the onboard cache is a DRAM volatile memory built by the manufacturer, and is used to cache write requests.

Further, the cache queue node refers to that when the upper-layer application request is sent to the solid-state disk cache through the operating system and the block device layer, the write request is divided into several logical pages of physical page size, and each cache node corresponds to a logical page. Pages, cache queues are used to cache write requests and service read and write requests for cache hits.

Further, the garbage collection detection writing mechanism refers to detecting whether the physical chip to which the logical page that is about to be written back is mapped to is doing garbage collection in the cache area, and if so, delays the writing back of the logical page, otherwise further Use the channel parallel write mechanism for conditional judgment.

Further, the channel parallel writing mechanism utilizes the data transmission parallel characteristic of the internal channel of the solid state disk to write back multiple logical pages belonging to different channels in the cache area at one time.

The solid-state disk cache management method based on garbage collection and channel parallelism combines the garbage collection detection writing mechanism and the channel parallel writing mechanism to reduce the cost of cache misses to improve cache efficiency. Write, which improves the overall write bandwidth of the SSD.

Compared with the prior art, the present invention has the following advantages and effects:

(1) On the basis of the LRU algorithm, the present invention adapts to the internal garbage collection characteristics of the solid state disk and proposes a garbage collection detection and writing mechanism, so as to avoid resource conflicts between the upper layer application request I/O and the internal garbage collection I/O of the solid state disk, It reduces the cost of cache misses and improves cache efficiency.

(2) Based on the logical page information provided by the garbage collection detection writing mechanism, combined with the parallel read and write characteristics between the internal channels of the solid state disk, the present invention proposes a channel parallel writing mechanism, which can be written back with a write operation time not exceeding the number of channels of the solid state disk Multiple logical pages hide a large number of write operations and improve the overall write performance of the SSD.

Description of drawings

Fig. 1 is an example diagram of channel parallel write-back mechanism in a solid-state disk cache method based on garbage collection and channel parallelism disclosed by the present invention;

FIG. 2 is a system architecture diagram of a solid state disk caching method based on garbage collection and channel parallelism disclosed by the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Example

As shown in Figure 1 and Figure 2, this embodiment discloses a solid-state disk caching method based on garbage collection and channel parallelism, which is applied in a solid-state disk storage system and uses garbage collection to detect and write mechanisms to fully consider the underlying garbage of solid-state disks. The problem of resource competition between recycling I/O and upper-layer application write request I/O hides the waiting time of garbage collection. Considering the parallel characteristics among the internal channels of the solid-state disk, use one write-back operation to write back multiple data pages with the largest number of channels at the same time, and use the time of one write operation to hide the time of multiple write operations. In addition, when the conditions of the channel parallel writing mechanism are not met, the logical pages are written back according to a predetermined priority order. At the same time, all data pages in the cache maintain the hot and cold attributes and dirty attributes, and use the priority write-back method to perform page replacement operations for data pages in the cache. Finally, a performance test is performed on this solid-state disk cache management method, and the impact of the present invention on the performance of the flash memory system is evaluated from multiple perspectives. The present invention sets a temperature attribute and a clean/dirty attribute for each logical page in the on-board cache DRAM of the solid-state disk, which are used to dynamically identify the hot and cold state and the clean/dirty state of the logical page, dynamically adjust the attribute value, and simultaneously adjust the attribute value Dynamic adjustment, which combines the garbage collection detection writing mechanism with the channel parallel writing mechanism, is mainly divided into three steps:

1) Adjustment of hot and cold attribute values and clean/dirty attribute values of logical pages. Modify the logical page data structure, add an integer temperature attribute, and set the default value to 0, indicating that the logical page is a cold logical page. When a logical page in the cache is hit, the temperature value of the logical page is incremented by 1, and the increment is stopped after the temperature value reaches 2. The logical page temperature value is divided into two levels, 0 and 1 for cold logical pages, and 2 for hot logical pages. Modify the logical page data structure and add a Boolean dirty attribute. The default value is set to false, indicating that the logical page is a clean page. The clean/dirty attribute is dynamically adjusted according to the mapping information. When the logical page is inconsistent with the valid data of the mapped physical page , the dirty attribute is adjusted to true, indicating that the logical page is a dirty page.

2) Trigger garbage collection detection write mechanism. When the onboard cache DRAM space is sufficient, the write request is directly written into the cache; when the onboard cache DRAM space is insufficient, a cache write-back operation needs to be triggered. When traversing the logical page in the cache queue, the garbage collection detection write mechanism is triggered to determine whether the logical page is a GC page (wherein, the GC page refers to the chip where the physical page mapped to the logical page is currently performing garbage collection operations. ), providing information basis for channel parallel writing mechanism.

3) Trigger channel parallel writing mechanism. When traversing the cache queue, according to the information provided by the garbage collection detection writing mechanism, first search the cache queue for cold dirty pages (logical pages) that do not exceed the total number of SSD channels, belong to different channels, and are not GC pages. When the found cold dirty pages When the number of pages is greater than 1, stop the search process, directly write back multiple found cold dirty pages, delete the first cold dirty page in the cache queue at the same time, and directly adjust the remaining cold dirty pages to clean pages. If the number of found cold and dirty pages is less than or equal to 1, then search for cold pages, hot pages, and GC pages in sequence in the order of multiple cold dirty pages>single cold page>single hot page>GC page, and any one of them is searched successfully in order Terminate directly, and finally write back the found logical page. At this time, the channel parallel writing mechanism is completed.

To sum up, a solid-state disk cache management method based on garbage collection and channel parallelism proposed in this embodiment is to consider the internal garbage collection I/O resource occupation of the solid-state disk in the cache queue in the onboard cache DRAM of the solid-state disk Due to the parallel characteristics between the environment and the internal channels of the solid state disk, the garbage collection detection writing mechanism and the channel parallel writing mechanism are introduced to avoid I/O conflicts and write back multiple logical pages in the time of one write operation, hiding multiple write operations Time, making full use of the parallel write feature between SSD channels, reducing the cost of cache misses, improving the cache efficiency and write bandwidth of the SSD storage system, and improving overall performance.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (6)

1. A solid state disk buffer memory management method based on garbage collection and channel parallelism carries out write-back operation on a logic page in a solid state disk on-board buffer memory queue according to a garbage collection detection write mechanism and a channel parallel write mechanism which are preset, wherein I/O issued to a bottom layer is segmented into physical page sizes and then enters a buffer memory area, namely the logic page, and the method is characterized by comprising the following steps:

s1, after an upper layer application request is issued from a block device layer to a flash memory board cache by an operating system, maintaining a temperature attribute for each logic page for dynamically identifying the cold and hot states of the logic page, and maintaining a clean/dirty attribute for the logic page according to a mapping relation for dynamically identifying the clean/dirty states of the logic page; wherein, the regulation of the cold and hot states and the clean/dirty states of the logical pages is as follows:

when each logic page enters the cache, setting an integer temperature attribute for each logic page, marking a temperature state, and then adjusting the temperature attribute value of the logic page according to the cache hit times;

when each logic page enters the cache, setting a Boolean dirty attribute for each logic page, and marking the clean/dirty state of the logic page, wherein the logic page initially entering the cache is a clean page by default, and the dirty attribute state is adjusted according to the mapping relation;

s2, triggering cache write-back operation when the cache space of the solid disk is insufficient, and detecting whether a chip where the physical position mapped by each logic page is located is doing garbage collection operation by utilizing a garbage collection detection write mechanism when each logic page is traversed; the process is as follows:

s21, when a request arrives, firstly judging whether the request hits in a cache area;

s22, when the cache area fails to hit, judging whether the first cache insertion can be performed according to the residual cache space;

s23, when the cache space is insufficient, a cache write-back operation is required to be triggered, when a proper logic page is searched, a garbage collection detection write mechanism is triggered for each traversed logic page, and whether the solid-state disk chip mapped by the logic page is doing garbage collection operation or not is judged;

s3, when traversing the cache queue, according to information provided by a garbage collection detection writing mechanism, the cold/hot state and the clean/dirty attribute of the logic pages, selecting a plurality of logic pages belonging to different channels, the number of which is not more than N, by taking the number N of internal channels of the solid-state disk as the upper limit, wherein the number is N, n=0, 1,2 and …, and when N is greater than 1, triggering the channel parallel writing operation to write back the N selected logic pages once; when n is less than or equal to 1, searching proper logic pages according to a preset priority order for writing back; the process is as follows:

s31, when a proper logic page is found, traversing is started from the tail part of the cache queue, whether the logic page is a GC page is judged according to information provided by a garbage collection detection writing mechanism, if yes, the next cache node is traversed directly, wherein the GC page means that a chip where a physical page mapped to the logic page is located is doing garbage collection operation;

s32, when the traversed logical page is not a GC page, judging whether the logical page is a cold dirty page according to the temperature attribute value and the clean/dirty attribute value of each logical page in the cache, if the logical page is the cold dirty page, continuing judging whether a channel mapped by the logical page is repeated with a logical page mapping channel in a logical page set for write-back, if not, adding the logical page into the logical page set until the number of the logical pages in the logical page set is equal to the number of channels in the solid disk or traversing the whole cache queue;

s33, when the number of the elements of the logic page set to be written back is larger than 1, triggering a channel parallel writing mechanism, and writing back all logic pages in the set at one time;

and S34, when the number of the logical page set elements prepared for write-back is less than or equal to 1, sequentially searching the cold page, the hot page and the GC page according to the priority order, stopping searching when one is searched, and finally writing back the searched logical page.

2. The method for managing solid-state disk cache based on garbage collection and channel parallelism according to claim 1, wherein the priority is as follows: multiple cold dirty pages > single cold page > single hot page > GC page.

3. The method for managing solid-state disk cache based on garbage collection and channel parallelism according to claim 1, wherein the on-board cache is a DRAM volatile memory built in by a manufacturer for caching write requests.

4. The method for managing solid-state disk cache based on garbage collection and channel parallelism according to claim 1, wherein the cache queue node refers to a logical page with a plurality of physical page sizes when a write request issued to the solid-state disk cache by an operating system and a block device layer is split into an upper layer application request, each cache node corresponds to one logical page, and the cache queue is used for caching the write request and serving whether the read-write request has a cache hit condition.

5. The method for managing solid-state disk cache based on garbage collection and channel parallelism according to claim 1, wherein the garbage collection detection writing mechanism is to detect whether a physical chip mapped to a logical page to which a write-back is about to occur is doing garbage collection operation in a cache region, if so, the write-back of the logical page is delayed, otherwise, the channel parallelism writing mechanism is further used for making a condition determination.

6. The method for managing solid-state disk cache based on garbage collection and channel parallelism according to claim 1, wherein the channel parallelism writing mechanism uses the data transmission parallelism of the internal channel of the solid-state disk to write back a plurality of logical pages belonging to different channels in the cache area at one time.