CN-121979816-A - Storage control and synchronous management method applied to multiprocessor system

Abstract

The invention provides a storage control and synchronization management method applied to a multiprocessor system, which comprises the steps of receiving multiprocessor access events, establishing a snapshot data set containing a processor identification code, an access resource address, an operation type and a time stamp, constructing an interleaved access map based on the data set, representing each event by a node, representing the relationship of time and resource conflicts among the events, setting map weights by time difference and resource sharing rate, carrying out synchronization risk assessment according to the map, estimating the possibility of data competition, deadlock or false sharing, and when the risk reaches a preset threshold value, readjusting the cache submitting sequence, inserting delay or lock control instructions by an interleaved correction scheduling algorithm, dynamically updating scheduling and buffering resource allocation to reduce the occurrence rate of subsequent conflicts and improve the data consistency and access efficiency of the multiprocessor system.

Inventors

• WANG CHONGYU

Assignees

• 晋达半导体有限公司

Dates

Publication Date

20260505

Application Date

20251201

Claims (2)

1. 1. A method for controlling and synchronizing storage in a multiprocessor system, comprising: (A) A snapshot recording step of receiving a plurality of access event requests sent by a plurality of processors at the same time, and respectively establishing a snapshot record for each access event request, wherein each snapshot record at least comprises a processor identification code, an access resource address, an operation type and an actual execution timestamp so as to form a group of access event data sets with time correlation; (B) The method comprises the steps of establishing a data set for structural analysis based on an access event data set, wherein the data set comprises a plurality of nodes and at least one side, the nodes respectively refer to specific access event records extracted from the access event data set, each node has unique identification, each side is defined as an associated data record connecting any two nodes and is formed according to the execution sequence or resource use conflict condition between the corresponding two specific access events, the time difference and the resource sharing rate between the nodes are defined as at least one map weight parameter used for representing the correlation degree or the influence intensity between the nodes, and the data set with the nodes, the sides and the map weight parameter is generally defined as an interleaved access map; (C) A synchronization risk evaluation step of quantitatively analyzing the map weight parameters of the nodes and the edges based on the cross access map established by the data set and defining a synchronization risk value capable of representing the access synchronization security, wherein the synchronization risk value is obtained by calculating the time difference, the resource sharing rate and the relevance of the edges among the nodes and is used for estimating the possibility of generating data competition, deadlock or false sharing among the nodes, and dividing the obtained synchronization risk value into a plurality of synchronization warning levels according to at least one preset threshold condition for the subsequent synchronization adjustment strategy or program dynamic optimization flow, and (D) An interleaving correction scheduling step is carried out when the synchronous warning level corresponding to the synchronous risk value reaches or exceeds the preset threshold value condition, wherein the interleaving correction scheduling algorithm comprises at least any one of (D-1) a data submitting sequence adjusting action, a cache submitting sequence, a data writing back time point or a writing batch segmentation strategy which correspond to the nodes according to the relevance which is characterized by the edge in an interleaving access map, and a conflict relieving control action, wherein the delay control instruction, the resource locking granularity adjusting instruction or the mutually exclusive access limiting condition of a block level is inserted into the specific node operation which is judged to have synchronous conflict risk, and the D-3) an access parameter and resource configuration adjusting action, namely updating the scheduling priority sequence, the capacity allocation of a cache or a buffer area or the access window opening time sequence of a plurality of processors according to the map weight parameter which corresponds to the nodes or the edge, so as to achieve dynamic synchronous optimization control.
2. 2. The method of claim 1 wherein in the interleaving correction scheduling step, scheduling priority boosting, processor access window reduction and deferred commit point adjustment are performed together for higher risk level access events to reduce system synchronicity impact of high risk events.

Description

Storage control and synchronous management method applied to multiprocessor system Technical Field The present invention relates to a data access control technology of a multiprocessor system, and more particularly, to a storage control and synchronization management method applied in a multiprocessor system, which is used for improving data consistency of a storage device in a multiprocessor environment, reducing buffer contention, avoiding deadlock and resource blocking, and improving access delay and overall system performance. Background In a multiprocessor System (MPS), a storage device is required to process parallel access requests of multiple processors at the same time. Conventional methods mostly use a single lock to control a shared resource (global lock), or the master processor polls the access requests (master-worker polling) of the slave processors sequentially, or a fixed scheduling policy, to maintain data consistency. However, these methods have several limitations, including first, when multiple processors frequently access shared buffers, data contention (meaning that multiple threads or processors access shared resources at the same time and at least one is a write operation, resulting in uncertainty or erroneous results), deadlock (meaning that two or more processors or threads wait for each other to release resources, resulting in the system not being able to continue to execute), or false sharing (meaning that multiple processors access different variables, but share the same cache line, resulting in unnecessary cache coherency update, decreasing efficiency), resulting in data coherency errors, second, global lock or polling mechanism may cause resource blocking and delay under high parallel load, decreasing system performance, third, fixed scheduling policy cannot dynamically adjust buffer allocation and IO control according to real-time access mode, lacking targeted synchronous optimization means, fourth, the prior art also has difficulty in establishing a quantifiable analysis model for access behaviors of different processors, resulting in potential synchronous risk not being predicted and actively controlled, especially, for today's high-speed storage devices such as Solid State Disk (SSD), multiple processors are prone to causing further problem of cache coherency and IO blocking. Therefore, it is difficult to achieve both data accuracy, high performance and dynamic synchronization control in the prior art, and there is a need for improved methods and systems for improving the overall performance and reliability of the storage device in a multiprocessor environment. Disclosure of Invention An object of the present invention is to provide a storage control and synchronization management method for a multiprocessor system, which is capable of automatically adjusting scheduling, lock granularity and buffer resource allocation by creating a snapshot of a plurality of processor access events, constructing an interleaved access map, and analyzing synchronization risk according to the map, so as to reduce the occurrence rate of data contention, deadlock and false sharing, and improve data consistency and overall system performance. The invention provides a storage control and synchronization management method applied to a multiprocessor system, comprising (A) a snapshot recording step of receiving a plurality of access event requests sent by a plurality of processors at the same time and respectively establishing a snapshot record for each access event request, wherein each snapshot record at least comprises a processor identification code, an access resource address, an operation type and an actual execution timestamp to form a set of access event data sets with time correlation, (B) a map construction step of establishing a data set for structural analysis based on the access event data sets, wherein the data set comprises a plurality of nodes and at least one side, the nodes respectively refer to specific access event records extracted from the access event data sets, each node has unique identification, each side is defined as an associated data record connecting any two nodes, each side is formed according to an execution sequence or a resource utilization conflict condition between the corresponding two specific access events, and each time difference and a resource utilization conflict condition are defined as a set of time correlation, (C) a map construction step of establishing a data set for structural analysis based on the access event data sets, wherein the data set comprises a plurality of nodes and at least one side, each node has unique identification, each side is defined as a risk value of the risk profile for the risk profile analysis, and the risk profile is quantized by the risk profile between the nodes and the nodes, the method comprises the steps of (1) obtaining a synchronous guard level corresponding to a synchronous risk value, and (D) executing an interleavi