A kind of multi-core parallel system and its data processing method

Abstract

The invention provides a multi-core parallel system and a data processing method of the multi-core parallel system. The method comprises the steps that data in a serialization critical area are distributed to a plurality of processors in the multi-core parallel system according to a first rule and are stored; when any of the processors needs to process the data in the serialization critical area and does not have the data to be processed, the processor for storing the data to be processed is determined according to the first rule, and a serialization critical area processing request is generated and sent to the determined processor; the processor receiving the serialization critical area request processes carries out processing according to the received serialization critical area processing request; after processing is completed, a processing result is returned to the processor generating the serialization critical area processing request. According to the multi-core parallel system and the data processing method, the processing performance of the multi-core parallel system on the serialization critical area can be improved.

Description

technical field [0001] The invention relates to the field of computers, in particular to a multi-core parallel system and a data processing method thereof. Background technique [0002] In recent years, processor manufacturers have been limited by power consumption and temperature, and in order to maintain the continuous growth of high-performance computing, computers have developed towards multi-core computer architectures. In order to make full use of the multi-core architecture, the application program is split into multiple threads that can run independently on a single CPU, so that the designed program can be executed in parallel to improve the overall operating efficiency. Based on such a program system, in order to ensure the synchronization and data integrity of data processing, threads need to use locks to access the shared area serially. The shared area of ​​the shared data serialization operation guaranteed by the lock is called the serialization critical area. ...

AI Technical Summary

Problems solved by technology

[0005] However, these solutions will inevitably lead to very large data migration. For example, when a fixed large processor needs to serialize the shared data distributed on other small processors, the data has to be migrated. Then operate; but with the increase in the number of CPU cores and parallel programs, this swap-in and swap-out operation becomes more frequent, and the delay of data migration between CPU networks makes the processing time of the serialization key area longer ; And when the operation is completed, the large processor will write the processing result to the local cache *Ln, and the data consistency processing increases the burden between the CPU network; due to the delay of the internal bus, the remote cache (cache) data on the chip The access will also become a bottleneck; therefore, when the number of CPU cores / parallel programs increases in the existing scheme, the performance of the multi-core parallel system will decrease, which will affect the scalability of the multi-core parallel system

Images