Präsentation herunterladen
Die Präsentation wird geladen. Bitte warten
Veröffentlicht von:Etzel Zeichner Geändert vor über 9 Jahren
1
TUM in CrossGrid Role and Contribution Fakultät für Informatik der Technischen Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur 24.05.2004 http://wwwbode.cs.tum.edu/Par/tools/Fundings/CrossGrid.html
2
Main roles: Participation in Task 2.4 “Interactive and semiautomatic performance evaluation tools“ Implementation of the High Level Analysis Component within the Grid application performance analysis tool G-PM Task leader of Task 2.1 “Tools requirements definition“ (finished) Task leader of Task 2.5 “Integration, testing and refinement“ Additional roles: Member of the Internal Review Board Member of the Architecture Team Member of the Integration Team Deputy leader of WP 2 Role of TUM in Crossgrid 1 Fakultät für Informatik der Technischen Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur 24.05.2004
3
What is G-PM? G-PM is an on-line tool that allows application developers to measure, evaluate, and visualize the performance of Grid applications G-PM is a unique tool for computer scientists and Grid programmers It combines performance analysis of applications at multiple abstraction levels with the analysis of the Grid infrastructure HLAC = High Level Analysis Component PMC = Performance Measurement Component UIVC = User Interface / Visualization Component OCM-G = Grid application monitoring system Benchmarks (Task 2.3) PMC UIVC HLAC OCM-G (Task 3.3) G-PM 2 Fakultät für Informatik der Technischen Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur 24.05.2004 Structure of G-PM:
4
3 HLAC adds a layer for high-level data analysis to G-PM, which provides two major functionalities to the user: 1.It enables to combine and/or correlate performance measurement data from different sources. E.g.: measure the load imbalance by comparing an application's CPU usage on each node measure the portion of the maximum network bandwidth obtained by an application by comparing performance measurement data with benchmark data 2.It allows to measure application specific performance metrics. E.g: the time used by one iteration of a solver the response time of a specific request convergence rate of an interative solver These functionalities are offered via user-defined metrics What is the Purpose of HLAC? Fakultät für Informatik der Technischen Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur 24.05.2004
5
4 What are User-Defined Metrics? User-defined metrics are performance metrics specified by the user at run-time according to his/her needs often they are specific to the examined application User-defined metrics can be based on existing metrics and optional information from the application: occurance of important events (probes) in the application‘s execution assosiation between related events (using a virtual time) performance data computed by the application itself Fakultät für Informatik der Technischen Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur 24.05.2004 In G-PM user-defined metrics are supported by a Performance Metrics Specification Language (PMSL)
6
5 Main Achievements of TUM Fakultät für Informatik der Technischen Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur 24.05.2004 After the second project year, TUM has achieved: definition of the PMSL language, based on requirements and examples of useful metrics for the CrossGrid applications implementation of measurements of metrics defined via PMSL parser for PMSL: translation into internal representation simple optimizations evaluation of measurements (centrally in G-PM, distributed evaluation is work in progress) full integration of HLAC with G-PM and OCM-G full integration of G-PM into the autobuild and deployment process G-PM / HLAC has been used with most CrossGrid applications: Blood flow simulation (Task 1.1) Flooding simulation (Task 1.2) High energy physics neural network training (Task 1.3) (Air pollution is in progress)
7
6 Dissemination Fakultät für Informatik der Technischen Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur 24.05.2004 Selected Presentations: 2 nd AcrossGrids Conference, Nicosia, Cyprus, 2004 University of Siegen, Germany, 2003 APART Workshop at EuroPar 2003, Klagenfurt, Austria Workshop on Clusters and Computational Grids for Scientific Computing 2002, Chateau de Faberges-de-la-Tour, France Dagstuhl-Seminar “Performance Analysis and Distributed Computing“, Germany, 2002 Selected Publications: R. Wismüller, M. Bubak, W. Funika, and B. Balis. A Performance Analysis Tool for Interactive Applications on the Grid. Intl. Journal of High Performance Computing Applications, 18(3), August 2004. M. Bubak, W. Funika, and R. Wismüller. A Performance Analysis Tool for Interactive Grid Applications. In Performance Analysis and Grid Computing, pp. 161-173. Kluwer Academic Publishers, 2003.
Ähnliche Präsentationen
© 2024 SlidePlayer.org Inc.
All rights reserved.