WO2020042388A1

WO2020042388A1 - Method for improving sequential read performance of solid state drive

Info

Publication number: WO2020042388A1
Application number: PCT/CN2018/116814
Authority: WO
Inventors: 白琮
Original assignee: 苏州韦科韬信息技术有限公司
Priority date: 2018-09-01
Filing date: 2018-11-22
Publication date: 2020-03-05
Also published as: CN109271107A; LU101773B1

Abstract

A method for improving sequential read performance of a solid state drive, comprising: before a newly generated sequential read command is added into a splicing unit, determining whether the new sequential read command conforms to splicing rules of the existing splicing unit; if yes, directly adding the new sequential read command into the splicing unit; if not, refreshing the existing splicing unit, and generating a new splicing unit; adding the sequential read command into the new splicing unit. The method for improving sequential read performance of the solid state drive further comprises a timeout trigger module which refreshes the existing splicing unit when a splicing unit does not have a new read command added beyond a pre-determined time. A plurality of commands is read in parallel, and the read performance of the solid state drive is effectively improved.

Description

Method for improving sequential read performance of solid state hard disk

Technical field

The invention relates to a solid-state hard disk control technology, and in particular, to a method for improving sequential read performance of a solid-state hard disk.

Background technique

An SSD (Solid State Drives) controller is a device that connects a user's host and storage particles (for example: NAND flash memory). One end of the SSD controller is connected to the host and the other end is connected to the storage particles. A compatible host is provided at the end connected to the host. A host interface protocol module of the interface protocol, and a storage particle interface protocol module conforming to the storage particle interface protocol is provided at one end connected to the storage particle. Because the speed of data transmission between the two interface protocol modules exists

Traditional page mapping manages the mapping of logical addresses to physical addresses for each logical page (commonly 4KB). After the host write command is issued, it will cut the 4KB size according to the data size described by the write command and distribute it into several 4KB data pages. These data pages are written into NAND flash sequentially, and the page mapping table records each data. The physical address to which the page is written forms the mapping relationship from logical address to physical address. This mapping management method makes the size of the mapping table very large and requires a large off-chip.

Summary of the Invention

In view of the above defects, the purpose of the present invention is how to improve the probability of sequential read and parallel operation, and overall improve the performance of the solid state hard disk.

In order to solve the above problems, the present invention proposes a method for improving the sequential read performance of a solid-state hard disk, which is characterized in that a splicing unit and a command judgment module are added to the read operation, and the command judgment module judges continuously whether the logical address of the read command is input. Whether the commands within a certain interval are sequential read commands. If so, add the current sequential read request to the stitching unit, and stitch the read commands according to the written rules. After the stitching is completed, the read commands of the same stitching unit are read in parallel. Way to access the storage unit.

The method for improving the sequential read performance of the solid-state hard disk is characterized in that the newly generated sequential read command first determines whether the new sequential read command conforms to the existing splicing unit before joining the splicing unit, and if it is, then directly joins the splicing Unit, if not, refresh the existing splicing unit and generate a new splicing unit, and add the sequential read command to the new splicing unit.

The method for improving sequential read performance of a solid-state hard disk further includes a timeout trigger module. When a splicing unit exceeds a preset time and no new read command is added to the splicing unit, the existing splicing unit is refreshed and executed.

A solid state hard disk is characterized in that a splicing unit and a command judgment module are added during a read operation, and the command judgment module continuously judges whether a command within a certain interval belongs to a sequential read command according to whether the logical address of the read command is input, and if so, The current sequential read request is added to the splicing unit, and after the splicing is completed, the read command of the same splicing unit is accessed in parallel to the storage unit.

The solid-state hard disk is characterized in that the newly generated sequential read command judges whether the new sequential read command conforms to the existing splicing unit before joining the splicing unit, and if it is, it is directly added to the splicing unit. The existing splicing unit is refreshed and executed, and a new splicing unit is generated, and the sequential read command is added to the new splicing unit.

The solid state hard disk further includes a timeout trigger module. When a splicing unit exceeds a preset time and no new read command is added to the splicing unit, the existing splicing unit is refreshed and executed.

The beneficial effects of the present invention are as follows: simply judging whether the logical address of the read command is a sequential read command or not, by adding the splicing unit, the read command is spliced according to the writing rule, and multiple commands are read in parallel. , Which effectively improves sequential read performance.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic diagram of a write address allocation rule for continuous LPA;

FIG. 2 is an execution flowchart when the task is driven;

Figure 3 is an execution flowchart triggered by the timeout mechanism.

detailed description

In the following, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The specific implementation of the technical points described in the foregoing is unfolded in order.

In the sequential read scenario, the logical address LPA of the read request is continuous, and the magnitude of the read request is relatively large. The source address of the data being read on NandFlash is related to the write distribution rules of these data. Figure 1 is the write of continuous LPA The schematic diagram of the incoming address allocation rule can simply show the effect of the write allocation rule. This write allocation rule can trigger the parallel read operation Multi_PlaneRead with the greatest probability when reading the scene sequentially. LPA0-LPA7 is distributed to the Page of the same index Index of Blockn and Blockm. Blockn and Blockm belong to two different Planes under a Die, which conforms to the Parallel Multi_Plane rule. achieve.

For how to identify sequential read requests, the method of this patent is that the firmware system determines the LPAID of the read request. If the continuous LPA reaches a certain number, it is determined that the system is currently in the sequential read scenario, and these and subsequent subsequent LPA reads are read. Requests are marked for sequential reading.

In the sequential reading scenario, the process of splicing mechanism execution (specifically, splicing into Multi_Plane operation) is shown in Figure 2 and Figure 3. Figure 2 is a task-driven execution flowchart; task-driven and time-out mechanisms cooperate, and task-driven splicing mechanism processes execute input-dependent requests. The request is first parsed and judged according to the request type. If the current request is a sequential read request, the process is continued based on whether there is currently a splicing unit. When a splicing unit exists in the system, it indicates that the previous sequential read request is newly generated or assembled into a splicing unit. At this time, the input sequential read request determines whether to join the splicing unit according to whether it conforms to the Multi_Plane rule of the splicing unit. If the Multi_Plane splicing rule of the splicing unit is not met, the current splicing unit is refreshed (that is, the NandFlash terminal executes the Multi_PlaneRead operation corresponding to this splicing unit), and the sequential read input is requested to newly generate the corresponding splicing unit. Whenever the current sequential read request is added to the splicing unit, the system checks the splicing unit to confirm whether the updated splicing unit has reached the maximum capacity of the splicing unit. For example, the sequential read scenario of LPA0-LPA7. The read request of LPA0 newly generates Page0's Multi_Plane splicing unit. The read requests of LPA1- LPA7 and Page0 splicing unit conform to the Multi_Plane rule. They are added to the splicing unit. When LPA7 is added to the splicing unit set At this time, the splicing unit contains read requests for LPA0-LPA7, and the inspection of the splicing unit has found that it has reached the maximum capacity (see Figure 1 that the largest LPAID of Multi_PlaneRead for Blockn and BlockmPage0 can only support LPA7), and then the Refresh action.

The timeout mechanism can increase the flexibility of the sequential read request time, avoid long-term waiting of the splicing unit, and also prevent the splicing unit caused by the request delay from being refreshed and executed without forming a Multi_Plane operation. Considering the sequential read scenario without the timeout mechanism control, this module receives the sequential read requests of LPA0-LPA3 and combines them into a splicing unit (only read requests of LPA0-LPA3 are received at one time instead of LPA0-LPA7 at one time. The reason for the read request is the module's scheduling strategy and the saturation of the request.) There are two possible strategies for processing the splicing unit: ① refresh immediately, refresh the splicing unit after receiving the last sequential read request, that is, the last LPA3 is received Refresh the splicing unit after a sequential read request. At this time, the splicing unit actually only has the Signle_PlaneRead of BlocknPage0; ② wait until the next request drives the refresh of the splicing unit (the request and the splicing unit do not comply with the Multi_Plane splicing rule) or update (sequential read request joins the splicing unit ), That is, after receiving the last LPA3 sequential read request, the waiting time is unknown. When strategy ① is adopted, it is possible that the next batch of sequential read requests for LPA4-LPA7 will arrive later. The Multi_PlaneRead that could have performed Blockn, BlockmPage0 is decomposed into the Signle_PlaneRead of Block0Page0 and the Signle_PlaneRead of BlockmPage0, the same read of LPA0-LPA7, NandFlash The execution time of the end is extended. When strategy ② is adopted, because the waiting time is unknown, the LPA0-LPA3 read requests that could have been completed during the waiting period have been waiting, which is a serious waste of performance. This patent introduces a timeout mechanism control, and records the time point of the newly generated splicing unit at the same time. When the current splicing unit s waits for more than a certain time, it will be refreshed and executed immediately.

What has been disclosed above is only one embodiment of the present invention, and of course, the scope of the rights of the present invention cannot be limited by this. Those of ordinary skill in the art can understand all or part of the processes for implementing the above embodiments and make according to the claims of the present invention. Equivalent changes still belong to the scope covered by the present invention

Claims

A method for improving sequential read performance of a solid-state hard disk, which is characterized in that a splicing unit and a command judgment module are added in the read operation, and the command judgment module continuously judges whether a command within a certain interval belongs to a sequence according to whether the logical address of the read command input is continuous The read command, if it is, adds the current sequential read request to the splicing unit, and splices the read command according to the writing rules. After the splicing is completed, the read command of the same splicing unit is accessed in parallel to the storage unit.
The method for improving sequential read performance of a solid state hard disk according to claim 1, characterized in that the newly generated sequential read command determines whether the new sequential read command conforms to the existing stitching unit before joining the stitching unit, and if it is Add the splicing unit directly, if not, refresh the existing splicing unit and generate a new splicing unit, and add the sequential read command to the new splicing unit.
The method for improving sequential read performance of a solid-state hard disk according to claim 1, further comprising a timeout trigger module. When a splicing unit exceeds a preset time without a new read command being added to the splicing unit, the existing splicing unit is added. Refresh execution.
A solid state hard disk is characterized in that a splicing unit and a command judgment module are added during a read operation, and the command judgment module continuously judges whether a command within a certain interval belongs to a sequential read command according to whether the logical address of the read command is input, and if so, The current sequential read request is added to the splicing unit, and the read command is spliced according to the writing rules. After the splicing is completed, the read command of the same splicing unit is accessed in parallel to the storage unit.
The solid-state hard disk according to claim 4, wherein the newly generated sequential read command determines whether the new sequential read command conforms to the existing splicing unit before joining the splicing unit, and if so, directly adds the splicing unit, If not, the existing splicing unit is refreshed and executed, and a new splicing unit is generated, and a sequential read command is added to the new splicing unit.
The solid-state hard disk according to claim 4, further comprising a timeout trigger module, and when a splicing unit exceeds a preset time and no new read command is added to the splicing unit, the existing splicing unit is refreshed and executed.