Performance Analysis of Various Mechanisms for Inter-process Communication

Unix-based operating systems feature several forms of Inter-Process Communication (IPC) including pipes, Unix domain sockets, TCP/IP sockets, FIFOs, semaphores, and signals. Although these mechanisms provide similar services, each is designed differently with varying levels of performance. To better understand how these mechanisms work and why they perform the way they do, their implementations were analyzed. Pipes, Unix domain sockets, and TCP/IP sockets are commonly used mechanisms and are the focus of this project. Although they allow for communication between processes on the same machine, their performance in terms of throughput can vary significantly on the same platform. The analysis of the IPC mechanisms involved examining the kernel source code of the Linux operating system and developing benchmark utilities to measure their performance. Overall, Unix domain sockets have proven to be the fastest mechanism, in many cases performing over two times faster than pipes and TCP/IP sockets. On the Intel Pentium III/4/Xeon processor systems used, pipes performed better than TCP/IP sockets over loopback, but the contrary was true for systems using Sun’s UltraSPARC TI processor. This paper describes the tools developed and methods used to profile the IPC mechanisms.