Thread distribution method and system of distributed type file system

Abstract

The invention discloses a thread distribution method and system of a distributed type file system. The method includes the steps of obtaining a reading-writing request carrying a file descriptor; judging whether a thread corresponding to the file descriptor exits or not; if yes, considering that the request is not the first reading-writing request corresponding to the file descriptor and a corresponding thread has already been distributed for the request, and distributing the reading-writing request to the corresponding thread; if not, considering that the request is the first reading-writing request corresponding to the file descriptor, distributing an appropriate thread for the request from a multi-thread pool, and distributing the reading-writing request to the thread with the fewest management files in a balance chain table. It can be seen that the aims of balancing the numbers of requests processed by all the threads in the multi-thread pool and ensuring the high processing efficiency and performance stability of the system are achieved.

Description

technical field [0001] The invention relates to the technical field of thread allocation, in particular to a thread allocation method and system of a distributed file system. Background technique [0002] At present, multi-threaded messages in a distributed file system are usually distributed using a hash algorithm based on file descriptors. This algorithm has a flaw, that is, it is easy for all read and write request tasks to be sent to one or several threads. Forwarding, especially when the file descriptors are not continuous, in extreme cases, all files will be allocated to one thread, while other threads are always idle, so that all read and write request tasks are backlogged into a thread team, resulting in many Requests cannot be processed for a long time, resulting in low system processing efficiency; at the same time, all read and write request tasks are backlogged into a thread team, which makes system resources not fully utilized and easily leads to a sharp decline...

AI Technical Summary

Problems solved by technology

[0002] At present, multi-threaded messages in a distributed file system are usually distributed using a hash algorithm based on file descriptors. This algorithm has a flaw, that is, it is easy for all read and write request tasks to be sent to one or several threads. Forwarding, especially when the file descriptors are not continuous, in extreme cases, all files will be allocated to one thread, while other threads are always idle, so that all read and write request tasks are backlogged into a thread team, resulting in many Requests cannot be processed for a long time, resulting in low system processing efficiency; at the same time, all read and write request tasks are backlogged in a thread team, making system resources not fully utilized, which can easily lead to a sharp decline in the performance of the entire system

Images