RSS-based concurrent process data processing method

Abstract

The invention provides a RSS-based concurrent process data processing method and device. The concurrent degree of the processes can be maximized by dynamically starting the processes of which the number is the same as that of the RSS receiving queues, and resource waste is avoided. Meanwhile, the processes are bound with different CPU IDs respectively, the situation that when multiple processes run on the same CPU, mutual exclusion access to the same receiving queue can be guaranteed only by adopting an additional lock scheme is avoided, and meanwhile resource consumption caused when the processes are switched between CPUs is also avoided.

Description

technical field [0001] The invention relates to the technical field of computers, in particular to an RSS-based concurrent process data processing method. Background technique [0002] The receiver scaling (Receive Side Scaling, RSS) of multi-network card receiving queue technology is a method of placing multiple tuples carried by the IP header of the network card received message in different receiving queues through hash mapping, and the queues are passed through Interrupts are bound to different CPUs. In this way, multiple CPUs can process packets concurrently, improving packet processing efficiency. [0003] When multiple processes concurrently process messages, the number of processes started is usually through the configuration file, or according to the number of CPUs, fork out the corresponding number of processes. The number of processes has nothing to do with the number of RSS queues. At runtime, Linux dynamically schedules processes, so multiple processes may run ...

AI Technical Summary

Problems solved by technology

[0004] Disadvantage 1: When the number of processes is less than the number of RSS queues, it is impossible to read all the receiving queue data at the same time, and the highest concurrency cannot be guaranteed

And when the number of multiple processes is greater than the number of RSS, it will cause some processes to be unable to read the receiving queue and waste resources

[0005] Disadvantage 2: When the process is not bound to the CPU, the system schedules the process. When multiple processes may run on one CPU, mutual exclusion locks are required to ensure access to shared resources, and the processing of locks is reduced. Packet Processing Efficiency

[0006] Similarly, when the system performs multi-process scheduling, the process needs to be switched between multiple CPUs, and each time the process is loaded into the kernel, the process environment variable needs to be re-introduced, which also reduces the processing efficiency of the process.

Images