Abstract

A common performance optimization for a Server process is to send the reply to each request as early as possible, before final operations that are not in the critical path (such as buffer cleanup, state updates, logging and file updates). The operations after the reply form a ``second phase'' of service. This does not delay the current request from the client, but may delay succeeding requests. The net performance improvement depends on the number of clients at a server, its utilization, and the proportion of the total work which is placed in the second phase. This dependence is explored using analytic models that include an improved special approximation for two phases service in queueing networks, and layered queueing networks The result is an approximate analysis for large and complex client-server systems, with second phases.",