Questions
- How does non-determinism arise in MPI applications?
- What are the impacts of non-determinism in scientific executions?
Objectives
Learn about different communication methods in MPI.
-
Learn how non-determinism impacts on scientific applications with real examples.
Tracking Scientific Incorrectness with ANACIN-X
Scenario
Let us consider a scientist, in conducting a meticulous experiment, initiaties two separate runs of the identical codebase. And, the outcome from these runs diverge significantly, raising questions about the underlying factors influencing the code's behavior.
Different outputs at separate runs of same code
Tracking Scientific Incorectness
The scientist uses ANACIN-X to pinpoint discrepancies between runs by generating Event Graphs for each run and plotting the KDTS (Kernel Distance Time Series).
Different outputs at separate runs of same code
Identifying Relevant Source Code Location
ANACIN-X automatically identify abrupt spike in kernel distance. These spikes imply sudden changes in the underlying communication non-determinism across executions.
This information helps to extract relevant source-code location via callstack vertex labels.
Now, the scientist knows where to look for the root cause of discrepant results.
