Performance evaluation of multi-core, multi-threaded SIP proxy servers (SPS)

Process schedulers are part of the core functionality of an operating system (OS), and have been enhanced over the years to account for multiple cores in the processors and to support multi-threaded applications. In this study, we investigate the impact of the Linux scheduler's load-balancing algorithm on the performance of multi-threaded OpenSIPS (an open source SIP proxy server, SPS) running on a multi-core processor system. Linux uses the “completely fair scheduler” (CFS) scheduling policy and provides parameters specifically tunable in a multi-core environment. We conducted extensive experiments and analyzed the collected data to characterize the performance of SPS as a function of the number of CPU cores, the number of server threads and the call arrival rate. Based on our analysis, we show how to configure the various scheduler parameters as a function of the number of CPU cores to achieve a significant improvement in SPS performance. We further present a capacity planning model as a tool that service providers may use to obtain a first-order approximation of the capacity of their system that yields a good match to experimental results.