CoCa: A Parallelization Model for Real-Time Data Analysis in High-Energy Physics

Peter van der Stok

Appeared in: IEEE Concurrency, April-June, Vol. 7, Nr 2, pp 38-46, 1999.

ABSTRACT

The scale, complexity and performance requirements of future High Energy Physics experiments make the parallelization of the software a necessity. A programming model, Communication Capability (CoCa), is presented that allows parallelization of the software at several levels of granularity. HEP software needs to recombine related data and reject data on many processors on the basis of the result of one single processor. Consequently, the isolation and atomicity properties of database transactions are attractive properties of our model. The complexity induced by the characteristics of the parallelized software is for a large part removed by the proposed model. The model has a higher level of abstraction than message passing while performance is shown to be comparable with PVM and MPI.