A probe effect in concurrent programs
Software—Practice & Experience
Models for performance perturbation analysis
PADD '91 Proceedings of the 1991 ACM/ONR workshop on Parallel and distributed debugging
Binary wrapping: a technique for instrumenting object code
ACM SIGPLAN Notices
Debugging with the MAD environment
Parallel Computing - Special double issue on environment and tools for parallel scientific computing
Time, clocks, and the ordering of events in a distributed system
Communications of the ACM
MPI-The Complete Reference, Volume 1: The MPI Core
MPI-The Complete Reference, Volume 1: The MPI Core
SKaMPI: A Detailed, Accurate MPI Benchmark
Proceedings of the 5th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface
On-Line Monitoring Support in PVM and MPI
Proceedings of the 5th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface
Monitoring Strategies for Hypercube Systems
PDP '96 Proceedings of the 4th Euromicro Workshop on Parallel and Distributed Processing (PDP '96)
Trace-based parallel performance overhead compensation
HPCC'05 Proceedings of the First international conference on High Performance Computing and Communications
Models for on-the-fly compensation of measurement overhead in parallel performance profiling
Euro-Par'05 Proceedings of the 11th international Euro-Par conference on Parallel Processing
| Hi-index | 0.00 |
The activities testing and tuning of the software lifecycle are concerned with analyzing program executions. Such analysis relies on state information that is generated by monitoring tools during program runs. Unfortunately the monitor overhead imposes intrusion onto the observed program. The resulting influences are manifested as different temporal behavior and possible reordering of nondeterministic events, which is called the probe effect. Consequently correct analysis data requires to keep the perturbation a minimum, which defines the need for monitors with small overhead. Measuring the actual overhead of monitors for MPI programs can be done with the benchmarking suite SKaMPI. It's results serve as a main characteristic for the quality of the applied tool, and additionally increase the user's awareness of the monitoring crisis. Besides that, the measurements of SKaMPI can be used in correction algorithms, that remove the monitoring overhead from perturbed traces.