Performance Optimization and Modeling of Fine-Grained Irregular Communication in UPC

Table 5

Comparison of actual and predicted time usages when running 1000 time steps of the 2D heat equation solver. The same hardware characteristic parameters as in Table 4 are used.

THREADS	Partitioning
Mesh size
16		0.52	0.33	122.53	122.07
32		0.44	0.37	61.55	61.04
64		0.27	0.21	30.78	30.52
128		0.29	0.21	15.31	15.26
256		0.18	0.13	7.70	7.63
512		0.14	0.14	3.85	3.81
Mesh size
16		1.55	0.65	489.96	488.28
32		1.08	0.73	246.25	244.14
64		0.64	0.42	122.82	122.07
128		0.64	0.42	61.85	61.04
256		0.42	0.26	31.01	30.52
512		0.29	0.26	15.47	15.26