Performance optimization method for multi-server multi-buffer zone parallel packet sending

The invention provides a performance optimization method for multi-server multi-buffer zone parallel packet sending. Driving software is in charge of allocating buffer zones used for sending messages, a message buffer zone is required to be applied for each thread in an inner core, and the number of CPUs (central processing units) connected with a memory can be specified to be the number of the threads by virtue of parameters when a memory is applied, namely the local memory of a CPU 0 can be applied for the thread 0, and the local memory of a CPU1 can be applied for the thread 1. Interface library software binds each thread onto the CPU corresponding to the thread number the thread while the thread calls an API (application programming interface) interface used for sending messages for the first time, thus effectively avoiding the overhead for CPU scheduling as the CPU accesses a remote memory and a remote thread and improving multi-thread packet sending efficiency.