Using and Extrapolating ScalaTrace Workloads in a Simulated Replay Environment
ScalaTrace is a lossless communication and I/O tracing tool with the capability to replay traces as MPI programs. ScalaTrace is currently unable, however, to replay in a simulation environment, which would allow us to experiment with replaying traces on future extreme-scale systems. This talk first describes integration of ScalaTrace technology into the CODES exascale storage system simulation toolkit to enable ScalaTrace I/O traces to be replayed in a generic replay environment.
The second part of this talk introduces the extrapolation of ScalaTrace communication traces to enable generating extreme-scale workloads without actually having to run them. The extrapolation of communication events has been explored in previous versions of ScalaTrace, but newer versions have more complex compression routines, making the existing extrapolation algorithm inappropriate and in need of reworking.
Xiaoqing Luo is a second year masters student in North Carolina State University. Her research interests include parallel application tracing and extrapolation of parallel programs.