Congratulations to Tamas who has just passed his viva with (very) minor corrections :)! I'm sure an announcement about drinks with Dr Suto will follow shortly!
The University [of Edinburgh] has promoted Jane Hillston to a personal chair in stochastic modelling, in recognition of her work on PEPA and her work in developing the field of stochastic process algebra. She has the distinction of being the first woman to be promoted to a personal chair in the subject area of computer science at The University of Edinburgh.
Stephen forgot to mention that he also has good news of his own. The University has also recognised his excellent work on PEPA and decided to promote him to Reader.
i'm very happy to announce the availability of the "luna" computational cluster.
we have 8 dual processor nodes (called luna01, luna02, ... , luna08), each with two 3GHz processors and 2GB RAM. the interconnect is gigabit ethernet. the nodes run linux and are connected to the department's file space by NFS (so you can just ssh in as usual). all aesop members are welcome to use the cluster - it makes a good training ground for mpi applications before they are ported to the viking cluster for example.
the performance of the cluster is not yet optimised (initial experiments with jumbo packets etc. resulted in various operational failures), but we will continue to work on this.
A hypergraph is an extension of a graph data structure in which edges (hyperedges) are allowed to connect arbitrary, non-empty sets of vertices. Like graphs, hypergraphs can be used to represent the structure of many sparse irregular problems such as data dependencies in distributed databases and component connectivity in VLSI circuits. Hypergraphs may also be partitioned such that a cut metric (a function of the interconnect in a partition) is minimised subject to a load balancing criterion. Hypergraph cut metrics provide a more accurate model than graph partitioning in many cases of practical interest such as the row-wise decomposition of a sparse-matrix for parallel matrix-vector multiplication. Algorithms for serial hypergraph partitioning have been studied extensively and tool support exists. However, these are limited by the processing power and memory available on a single processor. Unlike graph partitioning, there have so far been no parallel formulations of the serial hypergraph partitioning algorithms. This talk will outline the state-of-the-art serial hypergraph partitioning algorithms and describe two parallel algorithm formulations based on these. The second algorithm will formally be shown to be scalable. The two algorithms have been experimentally evaluated on large hypergraphs (with O(107) vertices) from domains of performance modelling and VLSI circuit design against state-of-the-art serial partitioning tools. Where problem size became intractable by serial computation, a parallel graph partitioning tool provided approximate partitions for comparison.