The usual solution to increase the availability of a system is redundancy at system level, which means adding spare computers being able to replace each other in the case of a failure. But this solution suffers from two severe drawbacks

due to the coarse level of redundancy, many non-live critical components have to be duplicated, resulting in very high costs. 

the interconnection network between the redundant nodes is relativly slow, leading to long fail-over times. 

A solution to both problems is structural redundancy at the peripheral bus level. A requirement for redundancy at this level is the possibility to remove and insert adapter cards from and into the bus during runtime, a process widely known as hot-swapping.

Due to their excellent cost/performance ratio, clusters of PCs can be attractive high performance computing (HPC) platforms. However, their limited communication performance over standard LANs is restrictive for parallel applications. It is the goal of the INFINIBAND-Cluster project at LRR-TUM to overcome this performance gap by adopting the INFINIBAND interconnect technology. With INFINBAND-based distributed shared memory (DSM) and high performance communication characteristics (throughput in the range of hundreds of Mbytes/s and latencies in the range of a few microseconds), a cluster built on top of this interconnection technology is regarded as well suited for HPC.

In many modern computers the speed at which arithmetic operations may be performed is far greater then the memory access times. A common approach to 'hiding' this problem is the use of fast access caches, however such configurations have direct consequences for the design and implementation of numerical algorithms. Only through the use of data locality is it possible to make efficient use of the cache, but many algorithms do not exhibit this feature and therefore only reach a small percentage of the possible performance. In this project we shall investigate how to efficiently implement adaptive methods for the solution of partial differential equations so as to make the best use of cache architectures.

Numerical simulation of electrostatic and electromagnetic fields.

DAB is a digital technology offering considerable advantages over today's FM radio, both to listeners and broadcasters.