Performance Optimisation and Productivity
A Centre of Excellence in HPC
RankDLB V1
Version's name: RankDLB V1 ; a version of the RankDLB program.Repository: [home] and version downloads: [.zip] [.tar.gz] [.tar.bz2] [.tar]
Patterns and behaviours: Implemented best practices: Solving dynamic load balance problems ·
This version demonstrated the original problem. A group of MPI ranks process work items repeatedly. After every iteration completes, some information is exchanged among neighbouring ranks and the throughput, i.e. the total numner pf processed work items is reported. At some point, rank 0 starts adding more work items and it can be observed how the throughput decreases.
