Operating System Support for High-Speed Communication

Emerging network technologies such as ber-optic transmission facilities and Asynchronous Transfer Mode (ATM) hold the promise of delivering data rates on the order of Gbits/s between individual workstations on local and wide-area networks 14]. This order-of-magnitude increase in network capacity, combined with the explosive growth in microprocessor performance, enables a range of innovative new applications of distributed computing. Distributed multimedia, including real-time audio and video, and supercomputing on clusters of workstations are examples of such emerging applications. One important factor that could dramatically innuence the success of these new technologies is the degree to which operating systems can make these networking resources available to application programs. The role of an operating system (OS) is to mediate and multiplex the access of multiple application programs to the computing resources provided by the underlying hardware. Ideally, the operating system should not itself consume a signiicant share of these resources. Unfortunately, current operating systems are threatening to become the bottleneck in delivering input/output (I/O) data streams to application programs at high rates 5, 7, 22, 21]. In particular, data streams between applications on hosts connected by high speed networks suuer bandwidth degradation and added latency due to the operating system running on the hosts. This paper looks at the I/O bottleneck in operating systems, with particular focus on high-speed networking. We start by identifying the causes of this bottleneck, which are rooted in a mismatch of operating system behavior with the performance characteristics of modern computer hardware. Then, traditional approaches to supporting I/O in operating systems are re-evaluated in light of current hardware performance tradeoos. This re-evaluation gives rise to a set of novel techniques that eliminate the I/O bottleneck. The root cause of the OS I/O bottleneck is that speed improvements of main memory have lagged behind those of the central processing unit (CPU) and I/O devices during the past decade 6]. In state-of-the-art computer systems, the bandwidth of main memory is orders of magnitude lower than the bandwidth of the CPU, and the bandwidths of the fastest I/O devices approach that of main memory 1. The previously existing gap between memory and I/O bandwidth has almost closed, and a wide gap has opened between CPU and memory bandwidth, leaving memory as a potential bottleneck. To bridge the gap between CPU and memory speed in modern computers, system designers employ sophisticated cache systems. A cache exploits locality of reference in memory accesses …