Name
Abstract | ||
|---|---|---|
Data motion is a significant factor affecting runtime performance. Data-intensive applications are subject to the effects of data motion more so than other applications. This research uses abstract machine models to calculate runtime performance expectations for a geological simulation program. The models are based on the time to execute double-precision floating-point instructions and the time to load operands and store results for those operations. Two extremes of cache memory are considered and provide expected bounds for the programu0027s runtime performance. In one case, cache memory is unlimited; once a datum is read into cache, it is always available. In the other case, the cache is extremely limited and each datum is removed from cache memory after it is used once. A model for inter-process communication is also incorporated.This research shows that simple models can provide accurate expectations for runtime performance for some operations. For one operation studied herein the observed runtime performance only exceeded the expected upper bound in one case by 2%. The models can become less accurate if memory utilization is high. This may occur if only one core is used or if a change to an algorithm results in larger data structures. |
Year | Venue | Field |
|---|---|---|
2016 | SpringSim (HPS) | Cache-oblivious algorithm,Cache invalidation,Cache pollution,Cache,Computer science,CPU cache,Least frequently used,Parallel computing,Cache algorithms,Cache coloring |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 4 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| David Apostal | 1 | 2 | 1.43 |
| Sara Faraji Jalal Apostal | 2 | 0 | 0.34 |
| Ronald Marsh | 3 | 31 | 7.89 |
| Travis Desell | 4 | 116 | 18.56 |