Module distributed 

Distributed optimization using MPI for large-scale parallel computation

This module provides distributed optimization algorithms that can scale across multiple nodes using Message Passing Interface (MPI), enabling optimization of computationally expensive problems across compute clusters.

Modules

algorithms

Distributed optimization algorithms

Structs

CommunicationConfig

Communication optimization configuration

DistributedConfig

Configuration for distributed optimization

DistributedOptimizationContext

Distributed optimization context

DistributedStats

Performance statistics for distributed optimization

FaultToleranceConfig

Fault tolerance configuration

LoadBalancingConfig

Load balancing configuration

MockMPI

Minimal rank/size-only stub, for tests of logic that only reads MPIInterface::rank/MPIInterface::size (e.g. the crate-private WorkDistribution) and never actually exercises cross-rank communication.

ThreadRankMpi

An MPIInterface implementation that simulates size MPI ranks using real OS threads within a single process, so distributed algorithms (algorithms::DistributedDifferentialEvolution, algorithms::DistributedParticleSwarm) can be exercised end-to-end — with genuine cross-rank broadcast/gather/allreduce/send/recv — without an actual MPI runtime or multi-process setup. See spawn_ranks to construct and drive a full communicator.

WorkAssignment

Work assignment for a process

Enums

DistributionStrategy

Work distribution strategies

ReductionOp

Reduction operations for MPI

Traits

MPIInterface

MPI interface abstraction for distributed optimization

Functions

spawn_ranks

Build a size-rank simulated communicator and run body on each simulated rank in its own thread, returning each rank’s result in rank order. body receives a ready-to-use ThreadRankMpi handle; construct a DistributedOptimizationContext from it to drive a real algorithm.