Distop: A Low-Overhead Cluster Monitoring System.
Daniel Andresen, Nathan Schopf, Ethan Bowker, Timothy Bower · 2003
Current systems for managing workload on clusters of workstations, particularly those available for Linux-based (Beowulf) clusters, are typically based on traditional process-based, coarse-grained parallel and distributed programming. The DESPOT project is building a sophisticated thread-level resource-monitoring system for computational, storage and network resources [2]. The original implementation of DESPOT was based on SGI's Performance Co-Pilot (PCP) to facilitate the collection of performance monitoring data and to provide an API for the scheduling algorithm to retrieve the data. Unfortunately, the overhead of PCP and the infrastructure required to use PCP slowed down the performance of the DESPOT scheduling algorithms. In this paper we present an alternative to PCP which we call Distop. Distop was developed specically to satisfy the needs of the DESPOT project for low-overhead, ne-grained resource-monitoring tools for per-process network and other resource usage. We also present experimental results indicating the overhead of our system is minimal while providing accurate resource utilization data.