Solid State Drive Traffic Control Methods
By recording and restoring the processing results and execution location information of requests inside the solid-state drive, the blocking problem caused by the unavailability of target execution resources is solved, resource utilization and processing efficiency are improved, the overhead of repeated processing is reduced, and the continuity and stability of requests are ensured.
Patent Information
- Application Number
- CN202611578478.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-13
- Publication Date
- 2026-11-10
AI Technical Summary
In the resource allocation of the internal processing stage of solid-state drives, when the target execution resources are temporarily unavailable, it causes request blocking and affects subsequent processing stages. Furthermore, the existing solutions fail to effectively handle reserved but unused execution resources, resulting in resource waste and increased processing overhead.
When the target execution resources are unavailable, record the results and execution location information of the completed processing stages, release unused reserved execution resources, and restore the execution state of the request according to the recorded information after the resources are restored, so that the request can continue to be executed from the next processing stage. At the same time, retain the necessary execution context and delete the reconstructable context.
It reduces the impact of resource blocking on other requests, improves the utilization of execution resources and processing efficiency, reduces the overhead of repeated processing, and ensures the continuity and stability of request processing.
Smart Images

Figure CN123067799A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of solid-state drive (SSD) flow control technology, specifically to a flow control method for SSDs. Background Technology
[0002] In existing technologies, solid-state drives (SSDs) typically employ multi-queue, request scheduling, and resource allocation methods to manage input / output request traffic, thereby improving parallel access capabilities and reducing resource contention between different services. Chinese patent application CN113656141A discloses an input / output I / O request processing method and host device. Addressing the resource allocation problem between virtual and physical I / O queues in NVMe devices, it determines whether a virtual I / O queue is mapped to a physical I / O queue based on I / O request traffic, enabling heavily loaded virtual machines to utilize idle physical I / O resources, thus improving the resource waste and performance loss caused by fixed binding methods. This solution focuses on the dynamic allocation of physical I / O queue resources on the host side or in virtualization scenarios, solving the problem of queue resource matching between different request sources.
[0003] Chinese patent application publication CN115145493A discloses a resource allocation method based on a ZNS SSD system. This method stores I / O stream requests from different applications in corresponding I / O queues, allocates Chip / DIE resources according to the request intensity of each I / O queue, and combines a block-level mapping mechanism to handle read and write requests, thereby reducing interference between strong I / O streams and weak I / O streams. While this scheme can improve fairness among different I / O streams through resource isolation, its resource allocation still mainly revolves around the allocation relationship between different I / O streams and the underlying Chip / DIE. It lacks a specific mechanism for handling situations where a single I / O request has been partially processed within the SSD's internal execution chain, but the target execution resource for the next processing stage is temporarily unable to receive the request. This addresses the issue of how to handle the processing results obtained in the previous stage, the execution location, and the reserved but unused execution resources.
[0004] Within a solid-state drive (SSD), requests typically undergo multiple processing stages, including protocol parsing, address translation, caching or queuing, and media access. Each stage has limited execution resources with capacity and reception conditions. When a target execution resource is busy, has a full queue, or is temporarily unavailable and cannot continue receiving requests, allowing the request to continue occupying its reserved but unused execution resources can lock available resources, preventing other requests that could proceed to subsequent stages from acquiring the corresponding resources. This can cause localized congestion to propagate upstream along the internal execution chain. Conversely, directly canceling and re-executing the request may repeat completed processing stages, increasing internal processing overhead. Therefore, existing technology still needs to provide an SSD flow control scheme that can preserve the basis for request recovery, release unused reserved resources, and allow execution to continue from subsequent stages once the target execution resource is restored. Summary of the Invention
[0005] The purpose of this invention is to provide a flow control method for solid-state drives, thereby addressing some of the drawbacks and shortcomings pointed out in the background art.
[0006] The present invention adopts the following technical solution to solve the above-mentioned technical problems:
[0007] The system acquires input / output requests submitted by the host, processes the input / output requests in the internal execution chain of the solid-state drive according to preset processing stages, and records the processing results and execution location information of the completed processing stages. Before the input / output request sends a media operation command to the non-volatile storage medium to change the storage state, if the target execution resource of the next processing stage cannot receive the input / output request, the system retains the processing results and the execution location information, releases the reserved execution resources that have been allocated to the input / output request but have not yet been used, and allocates the reserved execution resources to other input / output requests that meet the execution conditions for entering the subsequent processing stages.
[0008] After the target execution resource is restored to a state where it can receive the input / output request, the execution state of the input / output request is restored according to the processing result and the execution location information, so that the input / output request can continue to be executed from the next processing stage.
[0009] Furthermore, before releasing the reserved execution resources, the execution context corresponding to the completed processing stage is retained, and the temporary context that can be regenerated based on the execution location information is deleted; after the target execution resources are restored to a state that can receive the input / output requests, the temporary context is reconstructed based on the execution location information, and the execution state of the input / output requests is restored.
[0010] Furthermore, when allocating the reserved execution resources to other input / output requests, the reserved execution resources are obtained by selecting input / output requests that do not use the target execution resources in the subsequent processing stages according to the subsequent processing stages of each other input / output request, so as to prevent the blockage caused by the target execution resources being unable to receive input / output requests from spreading to other processing stages.
[0011] Furthermore, after the target execution resource is restored to a state where it can receive the input / output request, the input / output request is made to enter the next processing stage, and the execution position of the current processing stage is released when the input / output request leaves the current processing stage; the release event of the execution position is used as the triggering condition for the upstream input / output request to obtain the execution position, so that the execution position is passed upstream level by level along the internal execution chain.
[0012] Furthermore, when deleting the temporary context, a data mapping version identifier corresponding to the execution location information is recorded; after the target execution resource is restored, the data mapping version identifier at the time of restoration is compared with the recorded data mapping version identifier. If the version identifiers match, the temporary context is reconstructed according to the execution location information; if the version identifiers do not match, the changed data mapping information is updated and the temporary context is reconstructed.
[0013] Furthermore, before releasing the reserved execution resources, depending on whether the information required to restore the input / output request can be re-obtained based on the execution location information, the execution context corresponding to the information that can be re-obtained is determined as the temporary context, and the execution context corresponding to the information that cannot be re-obtained based on the execution location information is retained.
[0014] Furthermore, after the target execution resource is restored, the temporary context is kept in a deleted state until a receiving position for the next processing stage of the input / output request is established in the target execution resource; after the receiving position is established, the temporary context is reconstructed according to the execution position information, and the reconstructed temporary context is delivered to the receiving position.
[0015] Furthermore, a transfer identifier corresponding to each release event of the execution position is generated, and the transfer identifier is associated with the upstream input / output request that obtains the execution position; after the upstream input / output request enters the current processing stage, the transfer identifier is set to an invalid state, so that the same release event corresponds to one execution position acquisition operation.
[0016] Furthermore, when the upstream input / output request no longer meets the execution conditions for entering the current processing stage before obtaining the execution position, the upstream input / output request that no longer meets the execution conditions is identified as the original upstream input / output request, the execution position is kept in a released state, and the triggering condition is transferred to the target upstream input / output request that meets the execution conditions.
[0017] Furthermore, after the triggering condition is transferred to the target upstream input / output request, the association between the triggering condition and the target upstream input / output request is maintained until the target upstream input / output request obtains the execution position or no longer satisfies the execution condition; during the period of maintaining the association, when the original upstream input / output request resumes to satisfy the execution condition, the triggering condition is not transferred back to the original upstream input / output request.
[0018] This invention processes input / output requests submitted by the host according to preset processing stages, and records the processing results and execution location information of completed processing stages. Before an input / output request sends a media operation command to the non-volatile storage medium to change its storage state, if the target execution resource of the next processing stage cannot receive the request, the aforementioned processing results and execution location information are retained. Simultaneously, allocated and unused reserved execution resources are released, and the released resources are allocated to other input / output requests that meet the conditions for entering subsequent processing stages. This reduces the invalid occupation of internal execution resources by blocked requests, allows available execution resources to participate in the processing of other requests in a timely manner, and reduces the impact of local resource unavailability on the execution process of other requests.
[0019] After the target execution resource is restored to an acceptable state, this invention restores the execution state of the input / output request based on the retained processing results and execution location information, enabling the request to continue execution from the next processing stage without repeating the completed processing stage. Thus, while maintaining the continuity of request processing and the recoverability of the execution state, it reduces the internal overhead caused by repeated processing and helps suppress the spread of blockages along the internal execution chain of the solid-state drive, improving execution resource utilization, input / output request processing efficiency, and operational stability. Attached Figure Description
[0020] Figure 1 This is a flowchart of the solid-state drive flow control process of the present invention.
[0021] Figure 2 This is a schematic diagram illustrating the release and reallocation of reserved execution resources when input / output requests are blocked in Embodiment 1 of the present invention.
[0022] Figure 3 This is a graph showing the change in storage usage before and after the temporary context is deleted in Embodiment 1 of the present invention.
[0023] Figure 4 This is a flowchart of the data mapping version verification and input / output request recovery process in Embodiment 1 of the present invention.
[0024] Figure 5 This is a schematic diagram showing the execution position being passed upstream along the internal execution chain in Embodiment 2 of the present invention.
[0025] Figure 6 This is a diagram showing the transmission time and cumulative transmission time of the three-level execution position in Embodiment 2 of the present invention.
[0026] Figure 7 This is a diagram illustrating the process of transferring identifier association and triggering condition in Embodiment 2 of the present invention. Detailed Implementation
[0027] As attached Figure 1 As shown, in one embodiment, the solid-state drive (SSD) receives input / output requests submitted by the host and processes them sequentially in its internal execution chain according to pre-set processing stages. The internal execution chain can consist of multiple interconnected processing stages, each undertaking a corresponding processing task during the input / output request execution process. For example, a processing stage may include at least some of the following stages: a request parsing stage, an address translation or data mapping processing stage, a request scheduling stage, a media operation command generation stage, a media operation command sending stage, and a completion processing stage. These processing stages are connected in a pre-set execution order. After an input / output request completes a processing stage, the SSD records the processing result of that stage and execution location information that characterizes the current execution location. The processing result is used to save the completed processing content. The processing result may include at least one of the following: request parameters, address information, data mapping information, scheduling information, or status information generated in the current processing stage and required for subsequent processing stages. The execution location information is used to determine the current execution status of the input / output request in the internal execution chain. It may include the stage identifier corresponding to the currently completed processing stage, the stage identifier corresponding to the next processing stage, or a status identifier that can uniquely identify the above processing stage, so that the input / output request has a basis for resumption when execution is paused in the future.
[0028] Before an input / output request sends a media operation command to the non-volatile storage medium to change its storage state, the solid-state drive (SSD) determines whether the target execution resource for the next processing stage can receive the input / output request. The target execution resource is the execution resource required for the input / output request to enter the next processing stage. It may include queue entries for storing pending requests, buffer space for storing request-related data or status information, status table entries for recording the request's execution status, processing entry points for corresponding processing modules that can accept new requests, or other internal resources that can restrict the input / output request from entering the next processing stage. The SSD can determine whether the target execution resource can receive the input / output request through its available status identifier, available queue entries, available buffer space, available status table entries, or request reception status. For example, if the target execution resource does not have an available queue entry or available processing entry point that can be allocated to the current input / output request, or if the target execution resource is in a state of paused request reception, it is determined that the target execution resource cannot receive the input / output request; if the target execution resource has a corresponding resource that can be allocated to the current input / output request and allows the input / output request to enter, it is determined that the target execution resource can receive the input / output request.
[0029] When the target execution resource is in a state where it cannot receive input / output requests, the current input / output request cannot proceed to the next processing stage. At this time, the solid-state drive (SSD) does not continue to occupy the reserved execution resources already allocated for the input / output request but not yet actually used. Instead, it retains the recorded processing results and execution location information and releases the corresponding reserved execution resources. Reserved execution resources are execution resources that have established an allocation relationship with the current input / output request but have not yet actually undertaken the corresponding processing task before the current input / output request enters the next processing stage. These can include pre-allocated queue entries, buffer spaces, request descriptor table entries, or other reallocatable internal resources. When releasing reserved execution resources, the mapping relationship between the reserved execution resource and the current input / output request can be broken, and the corresponding reserved execution resource can be restored to an allocatable state, for example, by adding the corresponding queue entry, buffer space, or request descriptor table entry to the corresponding available resource set. The recorded processing results and execution location information continue to be associated with the request identifier of the current input / output request so that the corresponding information can be obtained later based on the request identifier.
[0030] Released reserved execution resources can be re-allocated. The solid-state drive (SSD) determines which input / output requests meet the conditions for entering the subsequent processing stage from other pending input / output requests, and allocates the released reserved execution resources to the corresponding other input / output requests. Meeting the conditions for entering the subsequent processing stage can include at least the following: the processing stage currently required by the corresponding input / output request can be executed; the required execution resources match the released reserved execution resources; and the input / output request has no pre-existing dependencies that would prevent its continued execution. Therefore, when an input / output request stops execution because the target execution resource is temporarily unavailable, the resources already occupied by that input / output request but not yet processed can be made available to other input / output requests that meet the conditions for continued execution, thereby reducing the impact of local resource blockage on the processing of other input / output requests in the internal execution chain.
[0031] The solid-state drive (SSD) can also continuously monitor the status of the target execution resources. This monitoring can be achieved either by the corresponding resource management module generating status notifications when the target execution resource status changes, or by the SSD reading the availability status of the target execution resource according to pre-set check intervals. When the target execution resource recovers from a state where it cannot receive input / output requests to a state where it can receive such requests, the SSD determines the completed processing content based on previously retained processing results and determines the position where the input / output request was paused based on execution location information. Based on this, the execution status of the input / output request is restored, allowing it to continue execution from the next corresponding processing stage without repeating the completed processing stage. During execution status restoration, the retained processing results and execution location information can be read based on the request identifier associated with the input / output request, and the corresponding status can be re-associated with the currently available execution resource. Through this process, unused reserved execution resources can be released during periods when the target execution resource is temporarily unavailable, and the execution process of the original input / output request can be restored based on the retained information after the target execution resource is restored. This improves the utilization efficiency of the SSD's internal execution resources and reduces the continuous occupation of other input / output requests when a single processing stage experiences resource blocking.
[0032] In one implementation, before releasing reserved execution resources, the solid-state drive (SSD) differentiates the current execution state of input / output requests. The execution state can be represented by the execution context corresponding to the input / output request, which stores information necessary for the current input / output request to continue execution in the internal execution chain. Execution contexts corresponding to completed processing stages and still needed for direct use upon resumption of execution are retained by the SSD. Execution contexts that can be regenerated based on execution location information and the information already retained by the input / output request are identified as temporary contexts and deleted before releasing reserved execution resources. Deletion means releasing the occupancy relationship between the temporary context's internal storage space or execution resources and the current input / output request, allowing the corresponding space or resources to be reallocated, without deleting the request identifier, execution location information, and processing results that need to be retained to identify the input / output request and determine its resumption location. Thus, while retaining the information necessary for the resumption of input / output request execution, the occupation of internal storage space or execution resources during execution pauses can be reduced. Once the target execution resources are restored to a state capable of receiving input / output requests, the solid-state drive regenerates the deleted temporary context based on the previously retained execution location information, and restores the execution state of the input / output requests in combination with the retained execution context, enabling the input / output requests to continue into subsequent processing stages.
[0033] When determining which execution contexts can be deleted as temporary contexts, the decision can be based on whether the information needed to resume input / output requests can be retrieved again using the execution location information and the retained request information. For information that can be retrieved using the execution location information, the corresponding execution context is designated as a temporary context and deleted during the pause of input / output request execution. For example, physical address information, routing information, or queue status information required for subsequent processing stages that can be retrieved based on the current processing stage, the logical address information saved in the input / output request, and the currently valid data mapping information can be considered as regenerable information. For information that cannot be retrieved using the execution location information and the retained request information, the corresponding execution context is retained. For example, intermediate processing results generated by the previous processing stage and not re-determined by the current state can be retained as execution contexts that need to be preserved. In this way, the execution context can be divided into reproducible and persistently retained parts, thus avoiding the deletion of information necessary for resuming execution that cannot be retrieved.
[0034] When deleting a temporary context, a data mapping version identifier corresponding to the current execution location information is also recorded. The data mapping version identifier characterizes the data mapping state used when deleting the temporary context and serves as the basis for determining whether the data mapping has changed when rebuilding the temporary context. The data mapping version identifier can be generated by the mapping management module that maintains the data mapping relationship and updated when information changes that may affect the recovery of the corresponding input / output request. For different data mapping ranges, corresponding data mapping version identifiers can be maintained separately, ensuring that the recorded data mapping version identifiers correspond to the data mapping range involved in the current input / output request.
[0035] After the target resource is restored, the SSD obtains the data mapping version identifier corresponding to the restoration and compares it with the data mapping version identifier recorded when the temporary context was deleted. If the two data mapping version identifiers match, it indicates that the relevant data mapping state did not change during the temporary context deletion. In this case, the temporary context can be directly reconstructed based on the execution location information and previously retained request information. When reconstructing the temporary context, the next processing stage to be restored can be determined according to the execution location information. Then, the corresponding data mapping information and the request parameters retained in the input / output requests are read to reconstruct the address information, routing information, queue status information, or other reconstructable states required for the next processing stage. If the two data mapping version identifiers do not match, it indicates that the relevant data mapping state changed during the temporary context deletion. In this case, the changed data mapping information is first updated according to the data mapping state at the time of restoration, and then the temporary context is reconstructed based on the updated data mapping information, execution location information, and the request information retained in the input / output requests. By comparing the data mapping versions, it is possible to avoid restoring input / output requests based on expired data mapping information, thus ensuring that the reconstructed temporary context maintains consistency with the data mapping state at the time of restoration.
[0036] In another implementation, the temporary context is not immediately rebuilt after the target execution resource is restored. Instead, it remains in a deleted state until a receiving position for the next processing stage is established within the target execution resource for the corresponding input / output request. The receiving position carries the execution state and related context information required for the input / output request to enter the next processing stage. It can be implemented using an available queue item, request status table item, buffer location, or other resource location capable of receiving the input / output request within the target execution resource. When establishing the receiving position, a location matching the input / output request can be determined from the currently available locations within the target execution resource, and the determined location can be associated with the request identifier of the input / output request, thereby preventing the location from being simultaneously allocated to other input / output requests. Before the receiving position is established, even if the target execution resource has been restored, the temporary context remains in a deleted state to avoid prematurely occupying the resources used to store the temporary context before the input / output request can enter the next processing stage. After the receiving position is established, the temporary context is rebuilt based on the execution position information, and the rebuilt temporary context is delivered to the corresponding receiving position, allowing the input / output request to continue its subsequent processing based on the receiving position.
[0037] In one implementation, when released reserved execution resources need to be reallocated to other input / output requests, the solid-state drive (SSD) determines the resource allocation object based on the subsequent processing stage that each other input / output request needs to enter. The SSD can check the subsequent execution path corresponding to each other input / output request and determine whether its subsequent processing stage requires the use of target execution resources that cannot currently receive input / output requests. The subsequent execution path can be determined based on the request type of the input / output request, the current processing stage, and pre-set processing stage connection relationships. For input / output requests that do not use target execution resources in their subsequent processing stages, they can be identified as candidate allocation objects for reserved execution resources, and the released reserved execution resources can be allocated to the corresponding input / output request, enabling the corresponding input / output request to continue executing its subsequent processing. When multiple candidate allocation objects exist, an input / output request can be determined from the candidate allocation objects according to the request scheduling order, request priority, or waiting order pre-set by the SSD, and the corresponding reserved execution resources can be allocated to it.
[0038] By employing the resource allocation method described above, when the target execution resource is temporarily unable to receive input / output requests, the reassignment of released reserved execution resources to subsequent input / output requests that still require the target execution resource can be avoided. For input / output requests in subsequent processing stages that do not depend on the target execution resource, the released reserved execution resources can be used to continue the corresponding processing. Therefore, the impact of the target execution resource's inability to receive input / output requests can be limited to the execution path related to that target execution resource, reducing the possibility of the blockage spreading to other processing stages in the internal execution chain, and improving the utilization efficiency of the solid-state drive's internal execution resources for input / output requests unaffected by the blockage.
[0039] In one implementation, once the target execution resource recovers from a state where it cannot receive input / output requests to a state where it can receive them, the input / output requests that were previously suspended due to the unavailability of the target execution resource are allowed to enter the next processing stage. After entering the next processing stage, the input / output requests no longer occupy the execution position corresponding to the current processing stage, thus releasing the execution position of the current processing stage. An execution position represents a location within a processing stage of an internal execution chain that can be occupied by an input / output request to maintain the execution state of the current stage. It can be implemented by a queue item, status table item, request slot, or an equivalent occupancy status within the corresponding processing stage. The difference between an execution position and the aforementioned reserved execution resources is that an execution position represents a processing position already occupied by the current processing stage of an input / output request, while reserved execution resources are resources pre-allocated for subsequent execution of input / output requests but not yet actually used for corresponding processing.
[0040] For upstream input / output requests in the internal execution chain, the release event of the execution position can be used as the trigger condition for acquiring the corresponding execution position. When an execution position is released, the upstream input / output request corresponding to that position gains the opportunity to enter the current processing stage, and acquires the execution position when the execution conditions for entering the current processing stage are met. An upstream input / output request is an input / output request located in a processing stage preceding the current processing stage and subsequently needing to enter the current processing stage. Meeting the execution conditions for entering the current processing stage can include the upstream input / output request having completed the necessary processing of the previous processing stage, the current execution position still being released, and the absence of other execution constraints preventing the upstream input / output request from entering the current processing stage. After an upstream input / output request acquires an execution position and enters the current processing stage, it can release the execution position it previously occupied in the previous processing stage, thus forming a process of execution positions being passed upstream level by level along the internal execution chain. In this way, idle execution positions created by downstream input / output requests moving to subsequent processing stages can be sequentially provided to upstream input / output requests, allowing each processing stage in the internal execution chain to continue receiving input / output requests according to the order in which execution positions are released.
[0041] To prevent a single execution location release event from triggering multiple upstream input / output requests to acquire the same execution location, a transit identifier is generated for the corresponding release event when an execution location is released, with each release event corresponding to a single transit identifier. The transit identifier can be implemented using a sequence number, label, or status record that distinguishes different release events, and it has both valid and invalid states. After generating the transit identifier, it is associated with the upstream input / output request currently identified as the target of the execution location acquisition. This association can be recorded in the status table or scheduling status information corresponding to the execution location, and includes the transit identifier, the execution location identifier, and the request identifier of the input / output request that is the target of the execution location acquisition. During the period of valid association, the upstream input / output request acquires the execution location based on the corresponding release event. Once the upstream input / output request acquires the execution location and enters the current processing stage, the transit identifier corresponding to the release event is set to an invalid state. After the transit identifier becomes invalid, no other input / output request will be triggered to acquire the execution location based on the same release event. The status updates of the transit identifier and the execution location occupancy status can be completed through atomic state updates, mutual exclusion access, or equivalent methods that prevent multiple input / output requests from simultaneously confirming the acquisition of the same execution location. Therefore, a release event for one execution location can correspond to a valid execution location acquisition process, avoiding execution location allocation conflicts caused by multiple upstream input / output requests responding to the same release event.
[0042] In actual execution, the upstream input / output request associated with the transfer identifier may undergo an execution status change before acquiring an execution position. When the upstream input / output request no longer meets the execution conditions for entering the current processing stage, it is identified as the original upstream input / output request and its eligibility as the current execution position acquisition target is revoked. Execution status changes may include the original upstream input / output request entering a waiting state due to the temporary unavailability of other execution resources, changes in its preceding processing conditions, or other state changes that temporarily prevent it from entering the current processing stage. In this case, the already released execution position is not reset to an occupied state, but remains in a released state. Simultaneously, a target upstream input / output request that meets the execution conditions is identified from other upstream input / output requests waiting to enter the current processing stage, and the triggering condition formed by the corresponding release event is transferred to the target upstream input / output request, making the target upstream input / output request the new execution position acquisition target. The transfer of the triggering condition can be achieved by keeping the transfer identifier corresponding to the release event in a valid state and updating the request identifier associated with the transfer identifier from the original upstream input / output request's request identifier to the target upstream input / output request's request identifier, without generating a new release event. This prevents upstream input / output requests that no longer meet the conditions for continued execution from occupying the released execution location, and allows upstream input / output requests that meet the conditions for execution to continue using the execution location.
[0043] After the triggering condition is transferred to the target upstream input / output request, the association between the triggering condition and the target upstream input / output request is maintained. This association remains unchanged until the target upstream input / output request obtains an execution position, even if the original upstream input / output request re-satisfies the execution condition. If the target upstream input / output request obtains an execution position, the corresponding triggering condition completes the execution position transfer and the corresponding transfer flag is invalidated as described above. If the target upstream input / output request no longer satisfies the execution condition before obtaining an execution position, the association between the target upstream input / output request and the triggering condition can be terminated, and the execution position acquisition object can be re-determined based on the currently satisfying upstream input / output request. When re-determining the execution position acquisition object, the transfer flag corresponding to the current release event and still in a valid state can continue to be used, and the request flag associated with that transfer flag can be updated. During the period when the triggering condition and the target upstream input / output request remain associated, even if the original upstream input / output request recovers to a state that satisfies the execution condition, the triggering condition will not be transferred back to the original upstream input / output request; the original upstream input / output request can continue to wait for other subsequently generated execution position release events. This prevents the triggering conditions from repeatedly shifting between multiple upstream input and output requests, keeps the transmission relationship of execution positions clear, and reduces execution position contention caused by changes in the execution state of input and output requests.
[0044] Example 1:
[0045] In one implementation, the solid-state drive (SSD) receives input / output requests submitted by the host and processes them within an internal execution chain according to pre-defined processing stages. The internal execution chain may include at least some of the processing stages such as request parsing, logical address processing, data mapping query, request scheduling, media operation command generation, and non-volatile storage media access scheduling, with each stage sequentially executed in a predetermined order. After each processing stage is completed, the SSD records the processing result of that stage and execution position information characterizing the current execution position of the input / output request. The processing result may include at least one of the following: request parameters generated in the current processing stage and still needed in subsequent processing stages: address information, data mapping information, and scheduling status information. The execution position information may include the stage identifier corresponding to the completed processing stage and the stage identifier corresponding to the next processing stage.
[0046] like Figure 2 As shown, the internal execution chain forms a sequential processing path based on the order of request parsing, logical address processing, data mapping query, request scheduling, media command generation, and media access scheduling. Taking a typical execution process as an example, the solid-state drive receives a write input / output request. This process sequentially completes request parsing, logical address processing, and data mapping query. Once the data mapping query is complete, The data mapping information corresponding to the logical address to be written has been obtained. The next processing stage needs to... The request is sent to schedule resources to further formulate media operation commands for changing the storage state of the non-volatile storage media. The solid-state drive saves the completed data mapping query results as processing results and records them. The execution location information for the current data mapping query stage and the next processing stage, the request scheduling stage, has been completed. Figure 2 The general requests a scheduling phase with a state representation distinct from other normal processing phases, to reflect... The execution position relationship that is paused and waiting due to the unavailability of target execution resources before entering this stage.
[0047] exist Before entering the request scheduling phase, the solid-state drive checks whether the target execution resource corresponding to the next processing phase can receive the request. The target execution resource can be a queue item, request status table item, buffer space, or processing entry point used to carry requests to be scheduled during the request scheduling phase. In this embodiment, the available request queue item in the target execution resource is used as an example. When all available request queue items in the request scheduling phase are occupied by other input / output requests, the solid-state drive determines that the target execution resource cannot currently receive requests. .at this time, Maintain the paused state before entering the request scheduling phase, and do not send media operation commands that change the storage state to non-volatile storage media. Figure 2 The text further illustrates The correspondence between the target execution resources, i.e. The target execution resource cannot enter the request scheduling phase because there are currently no available queue items. This blocking only restricts... The current subsequent execution does not require continued use of the already allocated space. Other execution resources that are reserved but not yet actually used.
[0048] Solid-state drives are paused. When continuing execution, retain the recorded processing results and execution location information, and check if the results have already been processed. Reserved execution resources that have been allocated but not yet actually put into use. Reserved execution resources are those already allocated to... A distribution relationship has been established, but no commitment has yet been made. Internal resources for subsequent actual processing tasks. In this embodiment, reserved execution resources may include pre-allocated request descriptor table entries, internal buffer space, and releaseable queue resources corresponding to subsequent processing stages. The solid-state drive releases the aforementioned unused reserved execution resources from... The occupancy relationship between them is determined, and the released reserved execution resources are reassigned. For example... Figure 2 As shown, unused reserved execution resources are released and restored to an allocatable state, then allocated by the solid-state drive to other input / output requests that meet the conditions for entering the subsequent processing stage. This separates the partial blocking of the target execution resource from the resource usage process of other unblocked requests, avoiding... During the pause, internal resources that are not immediately needed continue to be occupied.
[0049] Before releasing the reserved execution resources, the solid-state drive further... The currently formed execution context is distinguished. The execution context is used to store... Continue execution with the necessary status information. Specifically, solid-state drives (SSDs) retain information already generated in the aforementioned processing stages that is still needed during resumption of execution but cannot be retrieved based on the execution location information. Information that can be retrieved based on the execution location information... The execution context regenerated from the request identifier, execution location information, and data mapping information valid at the time of recovery is then determined to be a temporary context and deleted. Therefore, in During the pause, only the context resources necessary for resuming execution and which cannot be retrieved again are continuously occupied; information that can be retrieved again does not continue to occupy the corresponding internal storage space.
[0050] Figure 3 Further illustrations show the process before and after deleting the temporary context. The capacity changes of related execution contexts. Before deleting the temporary context, The relevant execution context occupies a total of 4096 bytes, of which 1536 bytes are used to store processing results and request states that cannot be retrieved based on the execution location information, and the remaining 2560 bytes are used to store data mapping derivation information, subsequent routing status, and scheduling preparation information that can be retrieved during the recovery phase. Therefore, the 4096 bytes consist of a persistent 1536-byte portion and a rebuildable temporary portion of 2560 bytes. After deleting the temporary context, the SSD retains 1536 bytes of execution context and releases the internal storage space occupied by the 2560 bytes of temporary context. Figure 3 By comparing the context usage before and after deletion, it is shown that the space occupied by reconstructable information can be released in a timely manner without deleting the necessary information required for recovery.
[0051] To represent the degree of reduction in execution context usage after deleting a temporary context, the reduction in context usage can be determined based on the context usage before and after deletion. Let... express The reduction in context usage after deleting the temporary context, expressed as a percentage (%). Indicates before deleting the temporary context The total storage capacity occupied by the related execution context, in bytes; This represents the execution context space that needs to be retained even after the temporary context is deleted, expressed in bytes. Since the total space before deletion consists of the retained portion and the removable temporary portion, the remaining space after deleting the temporary portion is the retained portion. Therefore, the degree of reduction in context space usage is expressed as follows:
[0052] .
[0053] In this example, It is 4096 bytes. It is 1536 bytes, therefore:
[0054] .
[0055] As can be seen from the above calculations, the deleted temporary context has a capacity of 2560 bytes, accounting for 62.5% of the total capacity of the execution context before deletion. After deleting the temporary context, 37.5% of the original total capacity of the execution context is still retained. Figure 3 The two sets of bar charts showing the capacity relationship are consistent with the calculation results above, namely, the total capacity before deletion is 4096 bytes, and the persistent retention capacity after deletion is 1536 bytes. Therefore, while maintaining... During this suspension period, provided that the necessary information for resumption of execution is available. The storage footprint of the relevant execution context was reduced by 62.5%, and the freed-up internal storage space can be used for other input / output requests.
[0056] In deletion When deleting a temporary context, the solid-state drive (SSD) simultaneously records a data mapping version identifier corresponding to the current execution location information. This data mapping version identifier is provided by the mapping management module that maintains the data mapping relationship and is used to characterize the deletion of the temporary context. The data mapping state on which it depends. In this embodiment, a monotonically updated mapping version number is used as an example. Delete. In the temporary context, the corresponding data mapping version number is V152. The SSD will associate V152 with... The request identifier and execution location information are associated and stored. The version number is used for determination. Whether the relevant data mapping changes during the suspension period is not required to use a specific numbering method. Figure 4 The document uses the mapping version V152 as one of the bases for recovery after temporary context deletion, and further illustrates the sequential relationship between target execution resource recovery, establishment of receiving location, mapping version comparison, and temporary context reconstruction.
[0057] After the reserved execution resources are released, the SSD determines which requests can continue to be executed from the other pending input / output requests, and selects resource allocation objects based on the subsequent processing stages of each input / output request. Specifically, the SSD checks the subsequent execution path of the pending requests and excludes those that still need to pass through the currently unprocessable paths. The request for the target execution resource. For input / output requests that do not use the target execution resource in subsequent processing stages but currently meet the conditions for continuing execution, they are identified as candidate requests, and the resource acquisition object is then determined from the candidate requests according to a preset request scheduling order. In an exemplary scheduling window, the solid-state drive includes... Multiple pause requests, including those for the currently blocked target, released a total of 20 reserved request resources that had not yet been used. Of these, 16 resources were subsequently allocated to other input / output requests that did not pass through the currently blocked target execution resources and were actually put into subsequent processing. The remaining 4 resources remained in an allocatable state within the scheduling window. Figure 2 The sequence of events from reserved execution resources to allocable states and then to other input / output requests that meet the conditions is used to represent the process of releasing and reallocating the resource.
[0058] To characterize the actual reuse rate of the released resources within the current scheduling window, it can be determined based on the relationship between the number of resources already put back into use and the number of resources actually released. Let... This indicates the actual reuse rate of the reserved execution resources released in this round, expressed in % (%). This indicates the number of reserved execution resources actually released in this round due to the target execution resource being unable to receive the corresponding input / output requests; the unit is units. This represents the number of released reserved execution resources that have actually been reallocated to other input / output requests that meet the execution conditions and put into use, expressed in units of [number]. The resources that have been reallocated and actually put into processing originate from the reserved execution resources already released in this round; therefore, the actual reuse rate is expressed as:
[0059] .
[0060] In this example, It is 20. The value is 16, therefore:
[0061] .
[0062] The calculations above show that 16 of the 20 reserved execution resources released in this round were actually used by other input / output requests, corresponding to an actual reuse rate of 80%. Another 4 reserved execution resources, accounting for 20% of the total resources released in this round, have not yet been reallocated within the current scheduling window. This result indicates that most of the reserved execution resources released due to target execution resource blocking can participate in the processing of other input / output requests within the current scheduling window, thereby reducing the continuous idle resources caused by local blocking.
[0063] The solid-state drive continuously monitors the availability of the target execution resources. It is able to receive requests when the scheduling phase reappears. When resources are available, the SSD does not necessarily restore the deleted temporary context immediately, but first... Establish the receiving location for the next processing stage. The receiving location can be a newly released request queue entry from the request scheduling stage or its corresponding request status table entry. The solid-state drive (SSD) determines a suitable location from the currently available locations to receive the request. The location, and compare that location with The request identifier establishes an association. Even if the target execution resource has recovered from an unavailable state to an available state, the association is not established until the receiving location is fully established. The temporary context remains deleted to avoid [further issues]. It is not yet possible to preemptively reclaim context resources before actually entering the next processing stage. For example... Figure 4 As shown, after the target performs resource recovery, it first establishes the receiving position for the next processing stage, and then enters the data mapping version verification process to ensure that the timing of the temporary context reconstruction matches the actual receivable time. The resource status remains consistent.
[0064] After the receiving location is established, the solid-state drive reads the data mapping version identifier corresponding to the recovery process and compares it with the deleted data. The data mapping version identifier recorded in the temporary context is compared. Figure 4 This comparison process is set as a branch decision node in the recovery path. When the data mapping version at the time of recovery matches the recorded V152, it indicates... During the pause, the relevant data mapping state remained unchanged. The SSD directly used the currently valid data mapping information and reconstructed the temporary context based on the previously retained execution location information and request state. When the data mapping version read upon resumption has changed to V153, it indicates... During the pause, the relevant data mapping status changed. The SSD reads the updated data mapping information corresponding to V153, and then combines it with... The execution location information and the retained request state are used to regenerate the temporary context. This branching process avoids using invalid mapping information obtained before deleting the temporary context when the data mapping has changed.
[0065] After completing the temporary context reconstruction, such as Figure 4 As shown, different versions of the comparison branch are re-integrated into the temporary context reconstruction and delivery path. The solid-state drive delivers the reconstructed temporary context to the already established temporary context. The established receiving location is then restored in conjunction with the previously retained execution context. The execution status. The solid-state drive determines this based on the execution location information. The completed data mapping query phase and the next request scheduling phase enable... There's no need to re-execute completed request parsing, logical address processing, and data mapping queries; execution continues directly from the request scheduling phase. Once the corresponding conditions are met during the request scheduling phase, media operation commands are generated and subsequent operations are performed on the non-volatile storage medium. In this way, while the target execution resource is temporarily unable to receive input / output requests, unused reserved execution resources and reconstructable temporary contexts can be released, and these resources can be made available to other input / output requests that meet the conditions for continued execution. After the target execution resource recovers, the original input / output requests can be restored based on the processing results, execution location information, and the valid data mapping state at the time of recovery, thereby reducing the continuous impact of local resource congestion on other request processing processes within the solid-state drive's internal execution chain.
[0066] Example 2:
[0067] In one implementation, the solid-state drive (SSD) receives input / output requests submitted by the host and processes them sequentially in an internal execution chain according to pre-defined processing stages. The internal execution chain includes multiple interconnected processing stages, each with an execution position available for input / output requests. The execution position carries the execution state of the input / output request within its corresponding processing stage and can be implemented using request slots, queue entries, status table entries, or other location resources used to maintain the execution state of the current processing stage. When an input / output request completes its current processing stage and enters the next stage, it no longer needs to occupy the execution position of the current stage. The SSD sets the execution position to a released state, allowing subsequent input / output requests upstream in the internal execution chain that need to enter the current processing stage to acquire the execution position.
[0068] like Figure 5 As shown, taking a single exemplary execution process as an example, the consecutive processing stages in the internal execution chain are represented as S1, S2, S3, and S4, where S1 is upstream of S2, S2 is upstream of S3, and S3 is upstream of S4. S1 to S4 are only used to distinguish different processing stages in this exemplary execution process and do not limit the number of stages or specific processing functions in the internal execution chain. Initially, input / output request Q2 is located in S1, input / output request Q3 is located in S2, and input / output request Q4 is located in S3. Q4 remains in S3 because the target execution resource corresponding to S4 is temporarily unable to receive the request. When the target execution resource corresponding to S4 recovers to a state where it can receive Q4, Q4 moves from S3 to S4, releasing the execution position previously occupied in S3. Figure 5 The state relationship between Q4 migrating from S3 to S4 and S3 forming an idle position is used to indicate that after a downstream request moves to a subsequent processing stage, an execution position is formed that can be used by an upstream request.
[0069] After the solid-state drive (SSD) detects that an execution location in S3 has been released, it uses this release event as a trigger condition for an input / output request upstream of S3 to acquire that execution location. At this point, Q3 has completed the pre-processing required to enter S3, and there are no other execution constraints preventing Q3 from entering S3. Therefore, the SSD identifies Q3 as the target for acquiring that execution location. After Q3 acquires the free execution location in S3 and enters S3 from S2, it no longer occupies the original execution location in S2, thus making the corresponding execution location in S2 idle. Subsequently, Q2, located in S1 and meeting the conditions for entering S2, acquires the free execution location in S2 and enters S2 from S1. (The text repeats itself here.) Figure 5As shown, the release of execution position S3 allows Q3 to move downstream. The movement of Q3 further releases execution position S2, which in turn allows Q2 to move downstream. Therefore, from the perspective of the availability of execution positions, the idle execution positions are manifested as a step-by-step transfer from the position corresponding to S3 to the position corresponding to S2 and then to the upstream direction.
[0070] To prevent a single execution location release event from being reused by multiple upstream input / output requests, the solid-state drive (SSD) generates a unique transit identifier for each release event upon detecting an execution location release. This transit identifier can be implemented using a release event sequence number, a token identifier, or a state record that uniquely corresponds to the current release event. In this embodiment, release event E27 and its corresponding transit identifier T27 are used as an example. After generating T27, the SSD associates T27 with the upstream input / output request currently identified as the execution location acquisition target and records T27, the identifier of the released execution location, and the request identifier of the corresponding input / output request in the request state table. While T27 remains valid, only input / output requests validly associated with T27 are allowed to acquire the execution location based on E27. When an upstream input / output request associated with T27 successfully acquires the execution location and enters the corresponding processing stage, the SSD sets T27 to an invalid state and synchronously updates the occupancy status of the corresponding execution location, preventing the same release event from triggering other input / output requests to acquire the same execution location again.
[0071] During the process of propagating the execution position upstream level by level, the time elapsed for each level of execution position propagation can be determined using the release event timestamps and execution position acquisition timestamps recorded by the controller. Since the release of the execution position at the next level occurs after the upstream input / output request at the previous level successfully acquires the corresponding execution position and enters the next processing stage, the release events and position acquisition processes at each level are consecutively linked according to their execution sequence. Therefore, the total propagation time from the release of the first-level execution position to the completion of the execution position acquisition at the predetermined upstream level can be obtained by sequentially adding up the individual propagation times of each level. Let... This indicates the total time taken for the execution position to be continuously transmitted upstream, in microseconds; This indicates the actual number of levels continuously passed upstream from the current execution position; Indicates the current propagation level of the statistics; Indicates the first In a multi-level transfer, the time taken from the generation of a release event at the corresponding execution location to the successful acquisition of that execution location by the corresponding upstream input / output request, expressed in microseconds, is as follows:
[0072] .
[0073] All levels The release and acquisition times can be determined by the event timestamp register or scheduling record within the solid-state drive controller. For example... Figure 6 As shown, taking a three-level execution position continuous transfer process as an example, the single-level transfer time of level 1 is 2 microseconds, the single-level transfer time of level 2 is 3 microseconds, and the single-level transfer time of level 3 is 2 microseconds. The cumulative transfer time after level 1 is completed is 2 microseconds, the cumulative transfer time after level 2 is completed is 5 microseconds, and the cumulative transfer time after level 3 is completed is 7 microseconds. Therefore:
[0074] .
[0075] Figure 6 The central bar represents the single-level transfer time corresponding to the three-level transfer process, while the broken line represents the cumulative transfer time as the transfer level increases, with cumulative times of 2 microseconds, 5 microseconds, and 7 microseconds respectively. The cumulative value at the completion of the three-level transfer is consistent with the sum of the single-level transfer times. Since each new release event is formed after the upstream input / output request of the previous level obtains the corresponding execution position and completes the stage transition, the statistical times of each level are consecutive. Therefore, there is no case of repeated accumulation of the same time interval in this example. The 7 microseconds is used to describe the execution position transfer sequence in a specific operation process and does not limit each processing stage to a fixed transfer time.
[0076] In actual execution, upstream input / output requests associated with the transmission identifier may undergo execution state changes before the execution location is obtained. For example... Figure 7 As shown, after the execution location generates a release event E27, the solid-state drive (SSD) generates a valid transfer identifier T27 and first associates T27 with the upstream input / output request Q3. If Q3 still meets the execution conditions for entering the current processing stage before acquiring the execution location, Q3 acquires the corresponding execution location, and T27 is deactivated, ending the execution location acquisition process corresponding to E27. If Q3 no longer meets the execution conditions for entering the current processing stage before acquiring the execution location due to changes in the state of preceding dependencies, temporary unavailability of other necessary execution resources, or changes in other execution conditions, the SSD cancels Q3's eligibility as the current execution location acquisition target, but does not reset the released execution location to the occupied state, while keeping T27 in a valid state.
[0077] When Q3 no longer meets the execution conditions, the solid-state drive determines the target upstream input / output request that meets the execution conditions from other upstream input / output requests waiting to enter the current processing stage. For example... Figure 7As shown, taking Q5 satisfying the current execution condition as an example, the SSD disconnects the association between T27 and Q3 used to obtain the current execution position and transfers the trigger condition formed by E27 to Q5, thus establishing a new valid association between T27 and Q5. This transfer can be achieved by updating the request identifier corresponding to T27 in the request status table, without needing to regenerate a new execution position release event or a new transfer identifier. While T27 and Q5 remain associated, even if the original upstream input / output request Q3 recovers to satisfy the execution condition for entering the current processing stage, the SSD does not transfer the trigger condition corresponding to E27 back to Q3, but instead maintains Q5 as the object for obtaining the current execution position.
[0078] If Q5 successfully acquires the execution location, the execution location acquisition process corresponding to E27 is complete, and T27 is set to an invalid state. If Q5 no longer meets the execution conditions before acquiring the execution location, the solid-state drive terminates the current association between Q5 and T27, and re-determines a new execution location acquisition target from other upstream input / output requests that still meet the execution conditions at this time. Figure 7 The release event E27, the transfer flag T27, the Q3 execution condition judgment, the transfer of the trigger condition to Q5, and the acquisition of the execution position by Q5 are connected in the order of state changes, so that the transfer flag remains valid before the execution position is effectively acquired, and becomes invalid after a qualified upstream input / output request actually acquires the execution position.
[0079] To characterize the time range within which the transmitted identifier remains effectively associated with the upstream input / output request after the trigger condition transition, the association duration can be determined based on the controller timestamps at the time of association establishment and at the time of association termination. Let... This indicates the start timestamp recorded when T27 establishes an association with the upstream input / output request of the target, in microseconds; This indicates the termination timestamp recorded when the upstream input / output request of the target successfully obtains the execution position or loses its execution qualification, in microseconds; The duration of the association between the triggering condition and the upstream input / output request of the target, expressed in microseconds, is as follows:
[0080] .
[0081] like Figure 7 As shown, taking the association process after T27 is transferred to Q5 as an example, the controller records the start timestamp when T27 and Q5 establish a valid association. The execution time is 125 microseconds, and a termination timestamp is recorded when the execution position is successfully obtained in Q5. It is 128 microseconds, therefore:
[0082] .
[0083] Therefore, the effective correlation between T27 and Q5 lasts for 3 microseconds. Figure 7 The association duration range corresponds to 125 microseconds to 128 microseconds. Within this time range, even if Q3 meets the execution conditions again, T27 remains associated with Q5 and does not re-transfer to Q3. When Q5 obtains its execution position at the time corresponding to 128 microseconds, T27 becomes invalid, and E27 is no longer used to trigger other input / output requests to obtain the same execution position. If Q5 loses its execution eligibility during the association period, the timestamp recorded when Q5 loses its execution eligibility is used as the reference. And based on this, determine the actual retention time of the current association.
[0084] In this way, the execution position release event generated after the downstream input / output request enters the next processing stage can drive the upstream input / output request that meets the execution conditions to obtain the corresponding execution position. As the upstream input / output request continues to move to the subsequent processing stage, the idle execution position is passed upstream level by level along the internal execution chain. By setting a one-to-one transmission flag for the release event, and transferring the triggering condition to the new target upstream input / output request when the original acquisition object loses its execution qualification, while keeping the association from reverting to the original upstream input / output request during the validity period of the new association, the same release event can correspond to a valid execution position acquisition process, reducing the situation of multiple upstream input / output requests competing for the same execution position and the repeated transfer of triggering conditions.
Claims
1. A flow control method for a solid-state drive, characterized in that, include: The system acquires input / output requests submitted by the host, processes the input / output requests in the internal execution chain of the solid-state drive according to preset processing stages, and records the processing results and execution location information of the completed processing stages. Before the input / output request sends a media operation command to the non-volatile storage medium to change the storage state, if the target execution resource of the next processing stage cannot receive the input / output request, the processing result and the execution location information are retained, the reserved execution resources that have been allocated to the input / output request but have not yet been used are released, and the reserved execution resources are allocated to other input / output requests that meet the execution conditions for entering the subsequent processing stage. After the target execution resource is restored to a state where it can receive the input / output request, the execution state of the input / output request is restored according to the processing result and the execution location information, so that the input / output request can continue to be executed from the next processing stage.
2. The flow control method for a solid-state drive according to claim 1, characterized in that, Before releasing the reserved execution resources, retain the execution context corresponding to the completed processing stage, and delete the temporary context that can be regenerated based on the execution location information; After the target execution resource is restored to a state where it can receive the input / output request, the temporary context is reconstructed based on the execution location information, and the execution state of the input / output request is restored.
3. The flow control method for a solid-state drive according to claim 1, characterized in that, When allocating the reserved execution resources to other input / output requests, the reserved execution resources are obtained by selecting input / output requests that do not use the target execution resources in the subsequent processing stages according to the subsequent processing stages of each other input / output request, so as to prevent the blockage caused by the target execution resources being unable to receive input / output requests from spreading to other processing stages.
4. The flow control method for a solid-state drive according to claim 1, characterized in that, After the target execution resource is restored to a state where it can receive the input / output request, the input / output request is made to enter the next processing stage, and the execution position of the current processing stage is released when the input / output request leaves the current processing stage; the release event of the execution position is used as the triggering condition for the upstream input / output request to obtain the execution position, so that the execution position is passed upstream level by level along the internal execution chain.
5. The flow control method for a solid-state drive according to claim 2, characterized in that, When deleting the temporary context, a data mapping version identifier corresponding to the execution location information is recorded. After the target execution resource is restored, the data mapping version identifier at the time of restoration is compared with the recorded data mapping version identifier. If the version identifiers match, the temporary context is reconstructed according to the execution location information. If the version identifiers do not match, the changed data mapping information is updated and the temporary context is reconstructed.
6. The flow control method for a solid-state drive according to claim 2, characterized in that, Before releasing the reserved execution resources, the execution context corresponding to the information that can be re-obtained is determined as the temporary context, based on whether the information required to restore the input / output request can be re-obtained according to the execution location information, while the execution context corresponding to the information that cannot be re-obtained according to the execution location information is retained.
7. The flow control method for a solid-state drive according to claim 2, characterized in that, After the target execution resource is restored, the temporary context remains in a deleted state until a receiving position for the next processing stage of the input / output request is established in the target execution resource. After the receiving position is established, the temporary context is reconstructed according to the execution position information, and the reconstructed temporary context is delivered to the receiving position.
8. The flow control method for a solid-state drive according to claim 4, characterized in that, Generate a transfer identifier corresponding to the release event of the execution location, and associate the transfer identifier with the upstream input / output request that obtains the execution location; After the upstream input / output request enters the current processing stage, the transmission identifier is set to an invalid state, so that the same release event corresponds to one execution position acquisition operation.
9. The flow control method for a solid-state drive according to claim 4, characterized in that, When the upstream input / output request no longer meets the execution conditions for entering the current processing stage before obtaining the execution position, the upstream input / output request that no longer meets the execution conditions is identified as the original upstream input / output request, the execution position is kept in the released state, and the triggering condition is transferred to the target upstream input / output request that meets the execution conditions.
10. The flow control method for a solid-state drive according to claim 9, characterized in that, After the triggering condition is transferred to the target upstream input / output request, the association between the triggering condition and the target upstream input / output request is maintained until the target upstream input / output request obtains the execution position or no longer meets the execution condition; during the association maintenance period, when the original upstream input / output request resumes to meet the execution condition, the triggering condition is not transferred back to the original upstream input / output request.
Citation Information
Patent Citations
Input / output (I / O) request processing method and host equipment
CN113656141A
Resource allocation method based on ZNS SSD system
CN115145493A
