An Evaluation of Network Stack Parallelization Strategies in Modern Operating Systems

As technology trends push future microprocessors toward chip multiprocessor designs, operating system network stacks must be parallelized in order to keep pace with improvements in network bandwidth. There are two competing strategies for stack parallelization. Messageparallel network stacks treat messages (usually packets) as the fundamental unit of concurrency, whereas connection-parallel network stacks treat connections as the fundamental unit of concurrency. Practical implementations of connection-parallel stacks map operations to groups of connections and permit concurrent processing on independent connection groups, thus treating the group as the unit of concurrency. Connection-parallel stacks can use either locks or threads to serialize access to connection groups. This paper evaluates these parallel stack organizations using a modern operating system and chip multiprocessor hardware.