We propose a model of TCP performance that captures the behavior of a set of network paths with diverse characteristics. The model uses more parameters than others, but we show that each feature of the model is important for at least some paths. We show that the model is sufficient to describe the datasets we collected with acceptable accuracy. Finally, we show that the model's parameters can be estimated using simple, application-level measurements.

Empirical skills are playing an increasingly important role in the computing profession and our society. In addition to being problem-solvers and designers/engineers, computer scientists must also be capable experimenters in order to develop, test, and evaluate complex hardware and software systems. The widespread use of computers as tools for interdisciplinary research also demands a strong grounding in the scientific method.This panel is designed to promote discussion about innovative methods for integrating empirical skills within the traditional computer science curriculum. Each panelist will define a set of core empirical concepts and skills that they see as essential to computer scientists, with a brief rationale for each. In conjunction, they will describe classroom practices that serve to demonstrate the key concepts and/or develop skills they have identified. Sufficient time will be allocated for discussion and contributions from the audience.

We evaluate pathchar, a tool that infers the characteristics of links along an Internet path (latency, bandwidth, queue delays). Looking at two example paths, we identify circumstances where pathchar is likely to succeed, and develop techniques to improve the accuracy of pathchar's estimates and reduce the time it takes to generate them. The most successful of these techniques is a form of adaptive data collection that reduces the number of measurements pathchar needs by more than 90% for some links.

This paper describes an operating systems (OS) class that departs from more common approaches by introducing experimental design explicitly as part of the course material. Instead of implementing operating systems components or modifying existing operating systems, students conduct a series of experiments that measure the performance of system services and try to infer information about their implementation from the results. These experiments reinforce the OS concepts presented in lecture, and also develop a set of skills that students need for a wide range of activities: formulating and testing hypotheses, analyzing and presenting data, and writing.

No abstract available.

We develop a workload model based on observations of parallel computers at the San Diego Supercomputer Center and the Cornell Theory Center. This model gives us insight into the performance of strategies for scheduling moldable jobs on space-sharing parallel computers. We find that Adaptive Static Partitioning (ASP), which has been reported to work well for other workloads, does not perform as well as strategies that adapt better to system load. The best of the strategies we consider is one that explicitly reduces allocations when load is high (a variation of Sevcik's A+ strategy (1989)).

We consider policies for CPU load balancing in networks of workstations. We address the question of whether preemptive migration (migrating active processes) is necessary, or whether remote execution (migrating processes only at the time of birth) is sufficient for load balancing. We show that resolving this issue is strongly tied to understanding the process lifetime distribution. Our measurements indicate that the distribution of lifetimes for a UNIX process is Pareto (heavy-tailed), with a consistent functional form over a variety of workloads. We show how to apply this distribution to derive a preemptive migration policy that requires no hand-tuned parameters. We used a trace-driven simulation to show that our preemptive migration strategy is far more effective than remote execution, even when the memory transfer cost is high.

We measure the distribution of lifetimes for UNIX processes and propose a functional form that fits this distribution well. We use this functional form to derive a policy for preemptive migration, and then use a trace-driven simulator to compare our proposed policy with other preemptive migration policies, and with a non-preemptive load balancing strategy. We find that, contrary to previous reports, the performance benefits of preemptive migration are significantly greater than those of non-preemptive migration, even when the memory-transfer cost is high. Using a model of migration costs representative of current systems, we find that preemptive migration reduces the mean delay (queueing and migration) by 35 - 50%, compared to non-preemptive migration.

No abstract available.