Skip to content

MPI Centralized

auto* s = gempba::mp::create_scheduler(gempba::mp::scheduler_topology::CENTRALIZED);

Implements scheduler. An actively routing multiprocessing scheduler. Uses OpenMPI as the IPC transport.

The center (rank 0) is an active participant: workers send excess tasks to the center, and the center dispatches them to available workers. Every task handoff traverses the center twice — once inbound, once outbound. At small process counts this is negligible; at scale the center becomes a bottleneck.

For the full protocol walkthrough and comparison with the semi-centralized topology, see Scheduler Topologies.

Center responsibilities

  • Maintain a bounded task queue
  • Maintain the process status table (RUNNING, AVAILABLE)
  • Dispatch queued tasks to available workers
  • Broadcast the global best value

The center's task queue has a configurable capacity limit. When it is full, workers are signaled to stop sending tasks. When it falls below threshold, workers resume.

Worker responsibilities

  • Explore the search tree via the thread pool
  • Send excess tasks to the center when it is not full
  • Accept tasks dispatched by the center
  • Update the local best value when the center broadcasts a new global best

When to use

Benchmarking and comparison only. The centralized topology exists to reproduce the experimental results from the original paper. For production use, prefer MPI Semi-Centralized.