A performance model for fine-grain accesses in UPC
UPC's implicit communication and fine-grain programming style make application performance modeling a challenging task. The correspondence between remote references and communication events depends on the internals of the compiler and runtime system. This correspondence is often hidden from application developers. Aggressive optimizations allowed by the relaxed memory consistency model further blur this correspondence by transforming code structure. A modeling approach based on UPC platform benchmarking and code analysis is proposed. This approach abstracts a UPC platform according to its potential to apply a few common optimizations, then divides remote references in the application code into groups, based on a dependence analysis, that are amenable to each optimization. Each group is associated with a cost, obtained via benchmarking each potential optimization. The aggregated cost of these groups is the predicted cost of the application. Three simple UPC applications modeled using this approach usually yielded performance predictions within 15 percent of actual running times. © 2006 IEEE.
20th International Parallel and Distributed Processing Symposium, IPDPS 2006
A performance model for fine-grain accesses in UPC.
20th International Parallel and Distributed Processing Symposium, IPDPS 2006,
Retrieved from: https://digitalcommons.mtu.edu/michigantech-p/10653