research relevant to the design and application of high performance scientific computers. We test our ideas by designing, building, and using real systems. The systems we build are research prototypes; they are not intended to become products. There is a second research laboratory located in Palo Alto, the Systems Research Center (SRC). Other Digital research groups are located in Paris (PRL) and in Cambridge,

Network-of-Workstations technology is applied to the challenge of implementing very high performance workstations for Earth and space science applications. The Beowulf parallel workstation employs 16 PCbased processing modules integrated with multiple Ethernet networks. Large disk capacity and high disk to memory bandwidth is achieved through the use of a hard disk and controller for each processing module supporting up to 16 way concurrent accesses. The paper presents results from a series of experiments that measure the scaling characteristics of Beowulf in terms of communication bandwidth, file transfer rates, and processing performance. The evaluation includes a computational fluid dynamics code and an N-body gravitational simulation program. It is shown that the Beowulf architecture provides a new operating point in performance to cost for high performance workstations, especially for file transfers under favorable conditions. 1 INTRODUCTION Networks Of Workstations, or NOW [?] ...

research relevant to the design and application of high performance scientific computers. We test our ideas by designing, building, and using real systems. The systems we build are research prototypes; they are not intended to become products. There is a second research laboratory located in Palo Alto, the Systems Research Center (SRC). Other Digital research groups are located in Paris (PRL) and in Cambridge,

Caching in the World Wide Web currently follows a naive model, which assumes that resources are referenced many times between changes. The model also provides no way to update a cache entry if a resource does change, except by transferring the resource's entire new value. Several previous papers have proposed updating cache entries by transferring only the differences, or "delta," between the cached entry and the current value. In this paper,

The interconnection of local area networks is increasingly important, but little data are available on the characteristics of the aggregate traffic that LANs will be submitting to the interconnection media. In order to understand the interactions between LANs and the proposed interconnection networks (MANs, WANs, and BISDN networks), it is necessary to study the behavior of this external LAN traffic over many time scales – from milliseconds to hundreds of seconds. We present a high time-resolution hardware monitor for Ethernet LANs that avoids the shortcomings of previous monitoring tools, such as traffic burst clipping and timestamp jitter. Using data recorded by our monitor for several hundred million Ethernet packets, we present an overview of the short-range time correlations in external LAN traffic. Our analysis shows that LAN traffic is extremely bursty across time domains spanning six orders of magnitude. We compare this behavior with simple formal traffic models and employ the data in a trace-driven simulation of the LAN-BISDN interface proposed for the SMDS SM service. Our results suggest that the pronounced short-term traffic correlations, together with the extensive time regime of traffic burstiness, strongly influence the patterns of loss and delay induced by LAN interconnection. 1.

This paper examines the phenomenon of congestion in order to better understand the congestion management techniques that will be needed in high-speed, cell-based networks. The first step of this study is to use high time-resolution local area network (LAN) traffic data to explore the nature of LAN traffic variability. Then, we use the data for a trace-driven simulation of a connectionless service that provides LAN interconnection. The simulation allows us to characterize what congestion might look like in a high-speed, cell-based network. The most

UNIX Internet servers with an event-driven architecture often perform poorly under real workloads, even if they perform well under laboratory benchmarking conditions. We investigated the poor performance of event-driven servers. We found that the delays typical in wide-area networks cause busy servers to manage a large number of simultaneous connections. We also observed that the select system call implementation in most UNIX kernels scales poorly with the number of connections being managed by a process. The UNIX algorithm for allocating file descriptors also scales poorly. These algorithmic problems lead directly to the poor performance of event-driven servers. We implemented scalable versions of the select system call and the descriptor allocation algorithm. This led to an improvement of up to 58% in Web proxy and Web server throughput, and dramatically improved the scalability of the system.

Memory-System Design Considerations for Dynamically-Scheduled Microprocessors Keith Istvan Farkas Doctor of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto 1997 Dynamically-scheduled processors challenge hardware and software architects to develop designs that balance hardware complexity and compiler technology against performance targets. This dissertation presents a first thorough look at some of the issues introduced by this hardware complexity. The focus of the investigation of these issues is the register file and the other components of the data memory system. These components are: the lockup-free data cache, the stream buffers, and the interface to the lower levels of the memory system. The investigation is based on software models. These models incorporate the features of a dynamically-scheduled processor that affect the design of the data-memory components. The models represent a balance between accuracy and generality, and ar...

We have studied the performance of a high-speed commercial spread-spectrum wireless LAN that uses the CSMA/CA multiple-access strategy. Employing synthetic workloads, we measured packet capture success more so than signal propagation characteristics. Specifically, we measured throughput, packet loss rates, range, and patterns of errors within packets. We conclude that CSMA/CA is quite successful in allocating bandwidth under stress, but that packet capture rate degrades very quickly once the LAN's effective range is exceeded. Hence, network maintainers should plan the layout of wireless networks at least as carefully as they plan wired networks. 1 Introduction Thanks to regulatory decisions and improvements in technology, the last several years have seen an explosion of developments in the commercial wireless sector of the telecommunications industry. So far the emphasis has been on low-speed, voice-oriented telephony applications. Another promising application is data movement over...

Wide-spread deployment of infrastructure WLANs has made Wi-Fi an integral part of today’s Internet access technology. Despite its crucial role in affecting end-to-end performance, past research has focused on MACprotocol enhancement, analysis and simulation-based performance evaluation without sufficient consideration for modeling inaccuracies stemming from inter-layer dependencies, including physical layer diversity, that significantly impact performance. We take a fresh look at IEEE 802.11 WLANs, and using a combination of experiment, simulation, and analysis demonstrate its surprisingly agile performance traits. Our main findings are two-fold. First, contention-based MACthroughput degrades gracefully under congested conditions, enabled by physical layer channel diversity that reduces the effective level of MACcontention. In contrast, fairness and jitter significantly degrade at a critical offered load. This duality obviates the need for link layer flow control for throughput improvement but necessitates traffic control for fairness and QoS. Second, TCP-over-WLAN achieves high throughput commensurate with that of wireline TCP under saturated conditions, challenging the widely held perception that TCP throughput fares poorly over WLANs when subject to heavy contention. We show that TCP-over-WLAN prowess is facilitated by the self-regulating actions of DCF and TCP congestion control that jointly drive the shared physical channel at an effective load of 2–3 wireless stations, even when the number of active stations is very large. Our results highlight subtle inter-layer dependencies including the mitigating influence of TCP-over-WLAN on dynamic rate shifting.