Analysis of TCP/IP for high performance parallel implementations

Abstract

This paper preseents an analysis of potential parallelism in TCP/IP is presented for two purposes. First, it is used as a basis for an implementation on a multiprocessing platform. Second, it reveals the areas where protocol modifications and/or tuning would yield more efficient implementations on multiprocessors. The focus is on the analysis, and detailed dependence graphs among the atomic TCP/IP protocol functions are presented. No specific implementation design is shown, since this highly depends on the particular hardware chosen.