Matthew I. Frank, Anant Agarwal, Mary K. Vernon
1997 Proceedings of the sixth ACM SIGPLAN symposium on Principles and practice of parallel programming - PPOPP '97  
Parallel algorithm designers need computational models that take first order system costs into account, but that are also simple enough to use in practice. This thesis describes the LoPC model, which is inspired by the LogP model, but which accounts for contention in message passing algorithms on a multiprocessor or network of workstations communicating via active messages. While LoPC is based on mean value analysis, it parameterizes architectures and algorithms in exactly the same way as the
more » ... gP model. LoPC takes the L, o and P parameters directly from the LogP model and uses them to predict the cost of contention, C, for processing resources. From LoPC's mean value analysis, which is itself straight forward, we derive several even simpler rules of thumb for common communication patterns. We show that the LoPC model can provide accurate predictions for client-server communication patterns and for algorithms with irregular, but homogeneous, communication patterns. In addition, we demonstrate how to adapt LoPC to deal with systems that include extra protocol processing hardware to implement coherent shared-memory abstractions.
doi:10.1145/263764.263803 dblp:conf/ppopp/FrankAV97 fatcat:hldl7ochq5ajdgtxvuipixqawe