Simplified workload characterization using unified prediction

Quantitative workload characterization is essential to high performance computer architecture design. Unfortunately, quantitative results are typically hard to interpret, reproduce and compare, due to the staggering amount of detail inherent in modern architecture. Source language, compiler technolo...

Full description

Saved in:
Bibliographic Details
Main Authors: Driesen, K., Colette, J., Feng Ji, Jourdain, M., Fahmi, M., Ghuneim, A., Hersi, E., Kahwa, J., Khan, H.R., Kwan, C., Mahyari, A., Miecknikowski, J., Oulmane, M., Perucic, M., Pirbay, A., Solomon, L., Taoko, J.J., Lip Hooi Tan, Feng Qian, Honghao Zhang, Lingyan Zhang, Su Zhang, Renner, W.
Format: Conference Proceeding
Language:English
Subjects:
Analytical models
Benchmark testing
Computational modeling
Computer aided instruction
Computer architecture
Computer science
High performance computing
Information analysis
Instruction sets
Poles and towers
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Quantitative workload characterization is essential to high performance computer architecture design. Unfortunately, quantitative results are typically hard to interpret, reproduce and compare, due to the staggering amount of detail inherent in modern architecture. Source language, compiler technology target ISA, and micro-architecture, intertwined with system aspects such as memory hierarchy and multitasking regime, all add to the complexity of workload characterization. We propose two simple metrics to characterize program execution: a footprint measures the "size" of a program, and a Unified Prediction profile shows its "complexity". These metrics are architecture-independent, and allow meaningful comparisons of program behavior at a numerical but abstract level. We believe they can provide direction to subsequent, more detailed and costly simulation efforts.
DOI:10.1109/ISPASS.2000.842296