MSc Thesis Proposal
Parallel and Interactive Applications on the Grid
The ARDA[1] working group at CERN is part of the European Unions EGEE (Enabling Grids for E-sciencE)[2] project which will help the European science community to make use of Grid Computing. Currently over 15000 computer nodes worldwide are made available to researchers everywhere around the globe.
In the context of ARDA project we study how to make the most efficient use of large computing resources provided by the Grid. In many scientific applications a given computational problem may be easily split in a large number of independent tasks which may be executed in parallel. This is the case, among many others, for data analysis in High Energy Physics (and similar data-mining applications), certain biomedical applications such as protein sequencing algorithms (BLAST) and Monte-Carlo simulations for medical physics.
We have been able to achieve significant optimization of these applications through DIANE framework for Master-Worker computing [3]. The framework provides a 'skeleton' [4] for the job execution, which takes care of load-balancing and automatic failure recovery. Similar strategies are used by Google in their search engine infrastructure [5]. The framework has been designed in a way which enables to plug-in a new type of application very easily (and without modifying the original application), similarly to the BOINC [6] project.
We are looking for students interested in large-scale, parallel Grid computing. Our current research activities focus on the following issues:
- support of complex workflows and other synchronization patterns (beyond trivial parallelism) with the underlying Master-Worker model,
- resource-sharing strategies in a community of the users of the Master-Worker framework,
- enabling of the interactive applications via a pool of reserved resources (Workers).
References
- http://arda.cern.ch
- http://public.eu-egee.org/
- http://cern.ch/diane
- http://homepages.inf.ed.ac.uk/mic/Skeletons
- https://db.usenix.org/publications/library/proceedings/osdi04/tech/dean.html
- http://boinc.berkeley.edu/papers.php