PERFORMANCE EVALUATION OF A WORKSTATION CLUSTER, TMC CM-5, AND INTEL PARAGON XP USING A PARALLEL HOMOLOGY ANALYSIS PROGRAM/

In the homology analysis, required memory size and computing time prop ortionally increase with the square of the number of amino acids in pr otein. In order to expand the limit of size and to carry out analysis effectively, we parallelized for homology analysis program adopting th e wavefront method with tiling. The performance of a workstation clust er, CM5, and Paragon was compared by using our program. For performanc e evaluation of these machines, we have made a model of our parallel h omology analysis computation. Using the model, we can understand the c haracteristics of these machines with respect to our program. Executio n time estimated by the execution model is in good agreement with the observation. The averaged error of the estimation to the observation o n a workstation cluster, CM5, and Paragon are 0.1%, 5%, and 2%. Almost linear speed-up has been observed from 1PU to 20PUs on them. The rati o of the performance of Paragon, CM5, and the workstation cluster is 1 :0.38:0.22, This ratio is constant to the variety of the size of probl em and the number of processors.