Tuning Performance via Metrics with Expectations

Modern server systems employ many features that are difficult to exploit by software developers. This paper calls for a new performance optimization approach that uses designated metrics with expected optimal values. A key insight is that expected values of these metrics are essential in order to ve...

Full description

Saved in:
Bibliographic Details
Published inIEEE computer architecture letters Vol. 18; no. 2; pp. 91 - 94
Main Authors Yasin, Ahmad, Mendelson, Avi, Ben-Asher, Yosi
Format Journal Article
LanguageEnglish
Published New York IEEE 01.07.2019
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects
Code tuning
Measurement
measurements
micro-architecture
Microprocessors
multi-core/single-chip multiprocessors
Multicore processing
Optimization
performance analysis
Servers
SIMD processors
Software development
Software development management
Tuning
Online AccessGet full text

Cover

More Information
Summary:Modern server systems employ many features that are difficult to exploit by software developers. This paper calls for a new performance optimization approach that uses designated metrics with expected optimal values. A key insight is that expected values of these metrics are essential in order to verify that no performance is wasted during incremental utilization of processor features. We define sample primary metrics for modern architectures and present three distinct techniques that help to determine their optimal values. Our preliminary results successfully provide 2x-4x extra speedup during tuning of commonly-used software optimizations on the matrix-multiply kernel. Additionally, our approach helped to identify counter-intuitive causes that hurt multicore scalability of an optimized deep-learning benchmark on a Cascade Lake server.
ISSN:1556-6056
1556-6064
DOI:10.1109/LCA.2019.2916408