The performance of distributed applications (such as file transfer, remote login, tele-conferencing, full-motion video, and scientific visualization) is influenced by several factors that interact in complex ways. In particular, application performance is significantly affected both by communication infrastructure factors and computing infrastructure factors. Several communication infrastructure factors include channel speed, bit-error rate, and congestion at intermediate switching nodes. Computing infrastructure factors include (among other things) both protocol processing activities (such as connection management, flow control, error detection, and retransmission) and general operating system factors (such as memory latency, CPU speed, interrupt and context switching overhead, process architecture, and message buffering). Due to a several orders of magnitude increase in network channel speed and an increase in application diversity, performance bottlenecks are shifting from the network factors to the transport system factors.
This paper defines an abstraction called an "Operating System Transport System Architecture" (OSTSA) that is used to classify the major components and services in the computing infrastructure. End-to-end network protocols such as TCP, TP4, VMTP, XTP, and Delta-t typically run on general-purpose computers, where they utilize various operating system resources such as processors, virtual memory, and network controllers. The OSTSA provides services that integrate these resources to support distributed applications running on local and wide area networks.
A taxonomy is presented to evaluate OSTSAs in terms of their support for protocol processing activities. We use this taxonomy to compare and contrast five general-purpose commercial and experimental operating systems including System V UNIX, BSD UNIX, the x-kernel, Choices, and Xinu.
