Bandwidth and latency guarantees in low-cost, high-performance networks
Andrew A. Chien, Jae Hwan Kim · 1997
Network service guarantees not only provide significant performance benefits to distributed computing systems (more balanced resource utilization, fast fault recovery, and fair network access), but they are also essential for many new applications requiring realtime communications with continuous data types (audio/video). Unfortunately, existing multicomputer network routers, which use simple arbitration techniques such as first-comefirst-serve or round-robin, have fundamental limitations in supporting these needs. Most algorithms that have been suggested for network service guarantees in long-haul, real-time communication networks are too complicated to be feasible in high-speed, low-cost switches for multicomputer networks. The simpler algorithms proposed, so far, provide only limited service guarantees or waste significant network resources. In this dissertation, we show that a variety of traffic types, both real-time communications as well as best-effort communications, can be supported efficiently in fast and inexpensive switches. After analyzing limitations of existing solutions, we present a novel, cost-effective resource control algorithm for service guarantees. Such cost-effective service guarantees not only provide substantial benefits to overall system performance, but can also further expand the domain of multicomputer applications to encompass distributed multimedia applications requiring guaranteed services.