Project Assignment 12
Continue cross-platform I/O and communication benchmarking and prediction
-
Re-predict MPI-IO collective operations using results from the OS cluster with the same
scheduling policy as on the HPC cluster (use "hostname" to verify the
process placement). Repeat the comparison between the interpolation, translation,
and scaling methods.
-
Perform prediction on the broadcast results with higher number of processor
number configurations (1, 2, 4, 6, 8, ... 16). Again, use the same scheduling
policy on the two clusters. Fix the error bar problem, and compare the three
prediction methods.
- Benchmark different array distributions of MPI collective operations with 16 processors,
and perform prediction using the OS cluster results as base curve. Requirements
are the same as the above two tasks.
Requirements
- For all benchmarks, schedule processors on the same chasis at the HPC cluster
- Use the machinefile to schedule processors at the OS cluster in the same way as the HPC cluster does.
- Plot standard deviation on the result curve.