Learning Automated Performance Analysis using PetaKit and the Bootable Cluster CD
By Ivan Babic, Mobeen Ludin, and Charlie Peck
Earlham College, Richmond Indiana
Based on the work of Sam Leeman-Munk
Shodor, Durham, North Carolina
This module teaches a method for evaluating the scalability of parallel programs. It includes a software toolchain, PetaKit, for automating the collection of performance data both for single runs and for parameter sweeps illustrating both strong and weak scaling. This is accomplished by providing a ready-to-run parallel and distributed operating environment, the Bootable Cluster CD (BCCD), with a variety of parallel programs which are already instrumented with PetaKit. The PetaKit software is also provided independently as part of the BCCD so that it can be incorporated into the student's own programs, either running under the BCCD or on any other Un*x platform. This combination supports exploration of:
- Speedup
- Efficiency
- Scaling - strong and weak
- Amdahl's law
- Gustafson-Barsis law
- Karp-Flatt metric
Upon completion of this module students should be able to:
- Run existing BCCD programs which are instrumented with PetaKit, collect and view the performance data.
- Instrument user or open source/community programs with PetaKit, collect and view the performance data.
-
Understand the basics of:
- Speedup
- Efficiency
- Scaling - strong and weak
- Amdahl's law
- Gustafson-Barsis law
- Karp-Flatt metric
The documents can be downloaded below.
Resources:
PetaKit Module Document (.pdf) : The module document in PDF format.
The Bootable Cluster CD : An archive of the code to be used with Bootable Cluster CD as enhancements for incorporating Performance Analysis using PetaKit.