Measurements of distributed operating systems

Message-based operating systems consisting of a small message-passing kernel supporting a collection of system-server processes are considered. For such an architecture to be practical, it is essential that basic messages be fast, since they often replace a kernel call in a traditional system. The design and implementation of such operating systems have been studied. It is shown that the limiting factor, especially for small messages, is typically not network bandwidth but processing overhead. A quantitative breakdown of the time spent in kernel into the different message-passing function is presented. It is found that a large percentage of the round-trip time can be attributed to short-term scheduling and control-block manipulation functions. Further, measurements of services times for typical services on Unix suggests that server computation times are comparable to communication times incurred in the message kernel. These results are useful for system architects, especially for evaluating hardware support for interprocess communication in such environments.<<ETX>>