Optimization algorithms and grid computing

Masters Defense by Alexander Jacobsen



The size of the optimization problems which different parts of the scientific community tries to solve has become so large and continues to grow, so that even with the fastest computers the solution can not be achieved within a suitable time horizon. One way to reduce this problem is to exploit the large amounts of processing power distributed systems such as the grid has.

However, it is far from all algorithms which lend themselves equally well to parallelization and distribution over a system with a communication time as slow as the grid proces.

In this thesis different types of parallelization methods and optimization algorithms will be analyzed for their applicability to grid computing. In addition I will construct a framework for the grid OneClick, which is part of the Minimum intrusion Grid (MiG). This framework aims at making it as easy as possible to use the grid, by managing the parallelization and distribution of the algorithms.

After creating the Framework, I came to the conclusion that OneClick has some limitations which makes it less suitable for solving large problems.

Censor: Christian Storm Pedersen, DAIMI

Supervisor: Brian Vinter, DIKU