close

The Department of Computer Science at the University of Cyprus cordially invites you to the PhD Defense entitled:

Grid Computational Resource Performance Ranking and Auditing

Speaker: Mr. George P. Tsouloupas
Affiliation: University of Cyprus, Cyprus
Category: PhD Defense
Location: Room 148, Faculty of Pure and Applied Sciences (FST-01), 1 University Avenue, 2109 Nicosia, Cyprus (directions)
Date: Monday, May 11th, 2009
Time: 13:00 - 14:00 EET
Host: Marios Dikaiakos (mdd AT cs.ucy.ac.cy)
URL: https://www.cs.ucy.ac.cy/colloquium/presentations.php#cs.ucy.pres.2009.tsouloupas

Abstract:
Over the recent years the area of Grid Computing has seen an astonishing growth. Grid infrastructures have become the platform of choice for large- scale eScience. The world's largest Grid infrastructure - EGEE - currently comprises 300 sites distributed around the world, petabytes of storage capacity and CPU's in excess of 80,000. The different computing resources in these heterogeneous infrastructures gather impressive and unprecedented computational potential, yet, in order to utilize them, users need mechanisms for selecting the right resources for the right job. Users and Virtual Organization administrators also need end-to-end mechanisms to evaluate the performance of resources and audit resources according to their advertised performance. This can be a complicated process, and when large infrastructures are involved, it becomes unmanageable and prohibitively tedious in the absence of specialized tools. Performance ranking in a large, shared, heterogeneous and dynamic environment is a complex task because it needs to be done in an efficient and unobtrusive way. At the same time, it has to address many different types of application that come from several Virtual Organizations. This thesis presents a methodology for putting correct, meaningful and contextualized performance information at the user's disposal, thus facilitating the ranking of computational resources based on customizable criteria. Contextualization is achieved by enriching the measurements with metadata about when, where, how and in many cases under what circumstances the measurement is obtained. The thesis proceeds to propose a userdriven approach for ranking resources by employing custom ranking functions. GridBench is an extensible tool that has been designed and implemented in the context of this thesis and along the lines of this methodology. It allows for context-augmented performance evaluation using several types of benchmarks, ranging from synthetic microbenchmarks to real-world parallel applications. It demonstrates how the evaluation and ranking process, an otherwise complicated and tedious task, can be simplified. Grid- Bench features a user-friendly graphical interface that facilitates the invocation of tests and benchmark and the collection, archival and analysis of results. A primary component, SiteRank, enables the interactive user-driven creation of custom ranking functions and provides a ranking of resources according to a users' specification. The methodology and tools are applied through several experiments to the largest production Grid infrastructure in existence today. Among the arguments of the thesis is that the use of evidence-based "measured" data, in contrast to the "quoted" data advertised in information services by resource owners, is imperative. The existing de facto approach for selecting resources according to performance is shown to be insufficient and unreliable.

Short Bio:
George P. Tsouloupas is a PhD Candidate at the Department of Computer Science, University of Cyprus.

  Other Presentations Web: https://www.cs.ucy.ac.cy/colloquium/presentations.php
  Colloquia Web: https://www.cs.ucy.ac.cy/colloquium/
  Calendar: http://testing.in.cs.ucy.ac.cy/louispap/XCS-3.0/schedule/cs.ucy.pres.2009.tsouloupas.ics