Performance Analysis Techniques for the Exascale Co-Design Process

Martin Schulz, Jim Belak, Abhinav Bhatele, Peer-Timo Bremer, Greg Bronevetsky, Marc Casas, Todd Gamblin, Katherine E. Isaacs, Ignacio Laguna, Joshua A. Levine, Valerio Pascucci, David Richards, Barry Rountree
PARCO
March, 2014

Efficient and effective performance analysis techniques are critical for the development of future generation systems. They are the drivers behind the required co-design process that helps establish the principles needed for their design. In this paper, we will highlight two such approaches: PAVE, a project that investigates mapping of performance data to more intuitive domains and uses advanced visualization techniques to expose problems, and GREMLIN, a system evaluation environment capable of emulating expected properties of exascale architectures on petascale machines. Combined with other approaches in system modeling and simulation, these projects enable us to provide a meaningful introspection into a target application's characteristics as well as its expected behavior and, more importantly, likely bottlenecks on future generation machines.

[DOI/EE link]

@inproceedings{SBBBBCGILPRR14,
  author    = {Martin Schulz and Jim Belak and Abhinav Bhatele and Peer-Timo Bremer and Greg Bronevetsky and Marc Casas and Todd Gamblin and Katherine E. Isaacs and Ignacio Laguna and Joshua A. Levine and Valerio Pascucci and David Richards and Barry Rountree},
  booktitle = {PARCO},
  day       = {26},
  ee        = {http://dx.doi.org/10.3233/978-1-61499-381-0-19},
  month     = {3},
  pages     = {19--32},
  publisher = {IOS Press},
  series    = {Advances in Parallel Computing},
  title     = {Performance Analysis Techniques for the Exascale Co-Design
Process},
  volume    = {25},
  year      = {2014}
}