epEBench: True Energy Benchmark

A4 Conference proceedings


Internal Authors/Editors


Publication Details

List of Authors: Simon Holmbacka, Robert Müller
Editors: Erwin Grosspietsch, Konrad Klöckner
Publication year: 2017
Publisher: IEEE
Book title: 25th Euromicro international conference on parallel, distributed and network-based processing
Start page: 426
End page: 429
ISBN: 9781509060597


Abstract

Current benchmark suites for evaluating energy
efficiency of modern computer systems fail to accurately enough
replicate real-world streaming applications. A reason for this is
that streaming applications (like multi-media applications) have
a very indeterministic load pattern which depends heavily on the
input data to the application. Partly, current benchmarks are
not replicating the instructions used for executing the workload
based on the input data used in such applications, and partly
benchmarks do not replicate the fluctuating workload level as a
result of the execution. Failing to replicate real-world situations
renders the results from energy benchmarking untrustworthy,
especially when executing streaming applications. This paper
presents epEBench – an opensource multi-core benchmark specifically
designed for evaluating energy efficiency. The benchmark
supports workload modeling and workload level manipulation to
generate realistic case studies not earlier supported by benchmark
tools. We show how the workload model of two real-world
applications has been extracted from their real execution and
how this information has been used to create workload models
for epEbench. We evaluate the accuracy of the currently created
workload models and demonstrate the benchmark suite.
Current benchmark suites for evaluating energy
efficiency of modern computer systems fail to accurately enough
replicate real-world streaming applications. A reason for this is
that streaming applications (like multi-media applications) have
a very indeterministic load pattern which depends heavily on the
input data to the application. Partly, current benchmarks are
not replicating the instructions used for executing the workload
based on the input data used in such applications, and partly
benchmarks do not replicate the fluctuating workload level as a
result of the execution. Failing to replicate real-world situations
renders the results from energy benchmarking untrustworthy,
especially when executing streaming applications. This paper
presents epEBench – an opensource multi-core benchmark specifically
designed for evaluating energy efficiency. The benchmark
supports workload modeling and workload level manipulation to
generate realistic case studies not earlier supported by benchmark
tools. We show how the workload model of two real-world
applications has been extracted from their real execution and
how this information has been used to create workload models
for epEbench. We evaluate the accuracy of the currently created
workload models and demonstrate the benchmark suite.

Last updated on 2019-17-10 at 04:16