University of Houston
Department of Computer Science

In partial fulfillment of the Requirements for the Degree of
Master of Science

GAUTAM PURANAM
will defend his thesis

Validating automatic compute node clustering using NAS Parallel Benchmarks


Abstract

Network topology information is important for the Grid scheduler to select appropriate resources. The Network Weather Service (NWS) is the most commonly used tool for monitoring the grid platform. Topology information is also needed to automatically and adaptively deploy NWS. Previously, based on the Internet geometric space model from Global Network Position (GNP), we use the latency between hosts to map the network topology into an N-dimensional geometric space. And based on the position of the nodes in this space, we automatically cluster the nodes into groups that represent different administrative domains. In order to effectively solve the problem of resource sharing and allocation we have to be able to find shared links inside each administrative domain. These links may become the bottleneck for communication between nodes. Since nodes inside same subnet are most likely to share the same links, it is very important that we group our hosts into two groups: intra-cluster and inter-cluster.

Machines in the same subnet or group will show significant interaction and similarities in response to events on the network. In our experiments we used TCP Ping to discover hosts that are part of any one of these groups. However, we found that the results with TCP Ping may be inaccurate and incomplete to model an intra-domain network because nodes may be subject to a higher system delay than transmission delay between a pair of intra-domain nodes.

In this thesis, we validate the results obtained from TCP Ping to accurately model nodes in the intra-domain network with extensive experiments using NAS Parallel Benchmarks.

Date: 27, April, 2005
Time: 9:00 AM
Place: 550-PGH

Faculty, students, and the general public are invited.
Thesis Advisor: Dr. Jaspal Subhlok