Performance Monitoring and Analysis for Wide-Area Distributed Applications

Wednesday, September 15, 1999 - 17:30
TH 331
Dan Gunter Lawrence Berkeley National Laboratory (LBNL)
The explanation for poor performance in a wide-area distributed application is frequently not obvious. Bottlenecks may occur at a host machine or in the network, or in an interaction between the two. The application's environment may also change dynamically during or between runs. Therefore, tools are needed to automatically gather, visualize, and analyze top-to-bottom and end-to-end monitoring information. This lecture presents a monitoring method as well as tools which help instrument and visualize system performance. Outline: I. Motivation II. Method III. NetLogger instrumentation - Summary of method - ULM format IV. Collaborators Pablo - SDDF - 3-D CAVE visualizations NWS V. Implemented tools NLV NLA Monitoring agents Demo VI. Future plans Java agents (reprise) NGI VII. Referenc

Dan Gunter graduated from SFSU with a M.S. in Computer Science in Spring, 1999. He worked as an intern at LBNL while completing his thesis, which is the subject of this presentation. Dan now works at LBNL full-time, fixing and creating bugs at an almost equal pace.