Medea: a tool for workload characterization of parallel systems

Abstract

Performance of parallel workload is heavily in uenced by h o w its own characteristics match the systems. Experimental approaches are particularly suitable, although di cult, for studying the behavior of the programs. The raw data collected by monitoring tools on real systems need to be processed for obtaining meaningful and easy{to{interpret information usable in identifying sources of poor performance. MEDEA (MEasurements Description, Evaluation and Analysis) is a software tool which meets such objectives. It has been designed as the natural extension of monitors in that it is typically used in strict conjunction with them. Various statistical and numerical techniques are applied to the collected data for deriving compact descriptions of the load, i.e., models, able to capture and reproduce its behavior. Visualization facilities are an integral part of our tool. As applications of MEDEA, a kernel and two real life programs (i.e., climate modelling and turbulent o w) running in di erent e n vironments are analyzed. Hints about possible code

Performance analysis of distributed applications using automatic classification of communication inefficiencies

“…In addition, numerous researchers have applied statistical techniques to performance data in an effort to reduce data volume or to automate tasks for the user. These techniques include covariance analysis, discriminant analysis, principle component analysis, and clustering analysis [3,19]. In the knowledge discovery field, Lee, Stolfo, and Mok [12] have focused techniques for machine learning on traces of Internet network activity to provide automated support for intrusion detection in computer networks.…”

A visualization tool for analyzing cluster performance data

“…These tools display system behavior through charts and graphs and do not show relationships between processors. A tool for workload characterizations is described in [4]. Analyzing system performance by visualizing a simulated system, described in [1], allows interactive layout of Gantt charts to display performance data.…”

Development and performance analysis of real‐world applications for distributed and parallel architectures

“…Based on the ALOG trace format we can use well-established visualization systems such as Medea [10], TAU [19], and Upshot [11]. Medea is a post mortem performance analysis and visualization system.…”