Towards Performance Predictive Application-Dependent Workload Characterization

Alkohlani, Waleed; Cook, Jeanine

doi:10.1109/sc.companion.2012.62

Cited by 6 publications

(1 citation statement)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Consequently, due to independency on microarchitectures, MICs are widely regarded as workload characteristics to represent program inherent behaviors. Traditionally, MICs are categorized into several groups consisting of instruction mix, inherent instruction level parallelism (ILP), register dependency distance, memory access locality, instruction locality and branch behaviors [5] [6] [7] [8] [9] [10].…”

Section: Introductionmentioning

confidence: 99%

Effect of Workload Characteristics on Similarity Analysis

Sha¹,

Xu²

2018

Proceedings of the 2018 International Conference on Computer Science, Electronics and Communication Engineering (CSECE 2018)

View full text Add to dashboard Cite

Workload characterization is the basis for similarity analysis, which is the core idea behind benchmark subsetting to pick up the most representative programs or program slices. The set of characteristics is crucial to the result of similarity analysis. Current studies typically use microarchitectureindependent characteristics (MICs) which reveal the inherent program behaviors to evaluate the similarities. In this paper, we propose a novel MICs: serializing instruction distance (SID). SID can describe the serializing instructions behavior that causes a significant performance loss of system-intensive mobile applications. The distribution of critical path length is also used as a MICs because it can reflect the inherent instruction level parallelism (ILP). Furthermore, we employ the comprehensive set of MICs to pick a representative set of program slices for each program of a mobile benchmark suites: Moby. The instructions per cycle (IPC) of each program slice is used to predict the whole program performance. The coefficient of variation of IPCs is under 6% and weighted average IPC prediction error is only 7%.

show abstract

Section: Introductionmentioning

confidence: 99%