parallel.md
In highs: 'HiGHS' Optimization Solver

Parallelism

HiGHS currently has limited opportunities for exploiting parallel computing. These are currently restricted to the dual simplex solver for LP, and the MIP solver. Details of these and future plans are set out below.

By default, when running in parallel, HiGHS will use half the available threads on a machine. This number can be modified by setting the value of the threads option.

By default, the HiGHS dual simplex solver runs in serial. However, it has a variant allowing concurrent processing. This variant is used when the parallel option is set "on", by specifying --parallel when running the executable via the command line, or by setting it via a library call in an application.

The concurrency used will be the value of simplex_max_concurrency. If this is fewer than the number of threads available, parallel performance may be less than anticipated.

The speed-up achieved using the dual simplex solver is normally bounded by the number of memory channels in the architecture, and typically less than the values achieved by Huangfu and Hall. This is because enhancements to the serial dual simplex solver in recent years have not been propagated to the parallel solver.

Unless an LP has significantly more variables than constraints, the parallel dual simplex solver is unlikely to be worth using.

The only parallel computation currently implemented in the MIP solver occurs when performing symmetry detection on the model, and when querying clique tables. This parallelism is always advantageous, so is performed regardless of the value of the parallel option.

The MIP solver has been written with parallel tree seach in mind, and it is hoped that this will be implemented before the end of 2024. The parallel LP solver will also enhance the MIP solver performance by speeding up the solution of the root node.

Development of a parallel interior point solver will start in 2023, and is expected to be completed by the end of 2024.