Predicting whole-program locality through reuse distance analysis

Abstract

Profiling can accurately analyze program behavior for select data inputs. We show that profiling can also predict program locality for inputs other than profiled ones. Here locality is defined by the distance of data reuse. Studying whole-program data reuse may reveal global patterns not apparent in short-distance reuses or local control flow. However, the analysis must meet two requirements to be useful. The first is efficiency. It needs to analyze all accesses to all data elements in full-size benchmarks and to measure distance of any length and in any required precision. The second is predication. Based on a few training runs, it needs to classify patterns as regular and irregular and, for regular ones, it should predict their (changing) behavior for other inputs. In this paper, we show that these goals are attainable through three techniques: approximate analysis of reuse distance (originally called LRU stack distance), pattern recognition, and distance-based sampling. When tested on 15 integer and floating-point programs from SPEC and other benchmark suites, our techniques predict with on average 94% accuracy for data inputs up to hundreds times larger than the training inputs. Based on these results, the paper discusses possible uses of this analysis.

HeapMD

“…2. As observed by several researchers, programs execute in phases [6,26,27,28], and different phases of the program exhibit different heap behavior. As the program phase changes, the heap-graph, and consequently some metrics associated with the heap-graph change to reflect the new heap behavior of the program.…”

Workload Modeling for Computer Systems Performance Evaluation

“…More efficient tree-based structures reduce the complexity from O(n m) to O(n log m) [666,22], where n is the length of the reference stream and m is the number of different addresses (note, however, that due to locality the cost of the simple algorithm is typically significantly lower than n m). Ding and Zhong give an approximate algorithm that reduces the time to O(n log log m) and the space to O(log m) [182].…”

“…Short reuse distances are used to characterize locality. But large reuse distances can be used to identify a return to an activity pattern that occurred a long time in the past (e.g., in a previous phase) [182].…”

CMcloud: Cloud Platform for Cost-Effective Offloading of Mobile Applications

“…To discover the application's temporal locality, we leverage the reuse distance (RD) analysis [10], in which RD is the number of distinct and different memory accesses between two consecutive references. Figure 5 shows an example reuse distance of four between two consecutive memory accesses to A.…”