用户态RPC协议分析及其多线程优化

RPC(Remote Procedure Call)协议有多个版本,分为内核态RPC与用户态RPC两类.开发人员根据设计需求选用合适的RPC版本,很多情况下会涉及跨内核态和用户态的通信.用户态RPC不如内核态RPC完善,缺少多线程机制、RDMA(Remote Direct Memory Access)机制等,需要优化以提高性能.结合此类需求,分析了用户态TI-RPC(Transport Independent Remote Procedure Call)运行机制,提出分层多线程的优化方法;利用TI-RPC底层接口重构RPC端口创建与服务启动;增加线程池机制,使TI-RPC在RPC层实现多线程并发服务.性能对比测试表明RPC内部多线程优化可将网络的利用率提高到网络满带宽的93%.

Analysis of User-Mode RPC Protocol and Multi-Thread Optimization

RPC (Remote Procedure Call) protocol has multiple versions, and can be roughly divided into two categories: kernel mode RPC and user-mode RPC. In the actual project environment, there are many cases that involves inter-state communication between kernel mode and user mode. The functions of kernel mode RPC is more complete than user mode RPC, such as multi-thread, RDMA(Remote Direct Memory Access). After analyzing the user-mode TI-RPC (Transport Independent Remote Procedure Call) operating mechanism under the above needs, hierarchical multithreading optimization on TI-RPC of user mode is proposed. The parts of the creation of TI-RPC ports and the start of services have been reconstructed by using bottom interface of the TI-RPC. With addition of thread-pool mechanism, the TI-RPC can achieve multi-threaded concurrent services on RPC layer. By comparing multi-threaded performance between RPC layer and RPC caller layer, the results show that the RPC internal multi-threaded optimization can improve the utilization of the network bandwidth to 93%.