Seminar „Praktikum Paralleles Rechnen“

aktuelle Informationen


Voraussetzung

Vorlesung "Paralleles Rechnen"

Zeit/Ort

Vorlesung:
Mittwoch, 13:30–16:45 Uhr, Raum 03.04.1.03

Start

29. 10. 2008

Umfang und Zuordnung

Leistungspunkte:
6 (4 SWS)
Informatikfachzuordnung:
Praktische Informatik (3000), Angewandte Informatik (4000)

Ansprechpartner

Anforderungen


Inhalt

In dem Seminar sollen aktuelle Themen aus dem Bereich Cluster und Grid Computing bearbeitet werden. Hierbei wird ein Schwerpunkt auf selbständige praktische Arbeiten gelegt.

Die Themen werden in Gruppen von 2-3 StudentenInnen bearbeitet. Die Themen kommen zum einen aus dem Bereich der Arbeitsgruppe Grid Computing von Prof. Schnor zum anderen aus Kooperation mit anderen wissenschaftlichen Einrichtungen in Potsdam, wie beispielsweise dem Potsdamer Institut für Klimafolgenforschung(PIK), dem Astrophysikalischen Institut Potsdam (AIP) und dem Albert-Einstein-Institut (AEI).

Weitere Informationen zum Seminar werden auf der Informationsveranstaltung (1. Vorlesung) gegeben.


Themen

1. Multicore Programming: OpenMP, MPI and UPC

Nowadays multicore architectures become more and more important, thus leading to these questions in the Parallel Computing field: Would the MPI-applications benefit from Clusters composed of multicore computation nodes? Or would the combination of OpenMP and MPI be more suitable for this architecture? How about the performance of Universal Parallel C (UPC) on multicore computers? This theme aims to answer these questions by implementing and testing the Cellular Automatons developed with pure MPI, pure OpenMP, UPC and the combination of OpenMP and MPI on multicore computers.

2. Chero as a Web Service

Chero is an application to solve a graphic theory problem in Bioinformatics. Since nowadays REST and WSRF/SOAP become the two main streams in Web Service field, this theme would allow students to develop Chero into a Web Service using both REST and WSRF/SOAP and integrate it with another service-based framework software Migol.

3. Sonivis & Terracotta

This topic deals with the parallelisation of an existing Java application (SONIVIS) using an existing Java clustering technique (Terracotta). The original Java code should stay unchanged as much as possible. The overhead of the parallelisation has to be measured.

4. libnetmsg: a lightweight communication library

Salbnet, is a load balance framework software from the current research work of our Cluster Computing workgroup, In order to make the load balance decision in Salbnet, a library is needed to send the credits of each backend-sever into the load balancer. Therefore, this theme is to develop such a lightweight, platform independent library to forward small-sized messages on UDP, TCP and InfiniBand.


Literatur

[zurück] [nach oben]
Mail to webmaster
© by Prof. Dr. Bettina Schnor, Klemens Kittan
Design by Klemens Kittan, Maximilian Haase