usage recommendations & start contributing

[ETROK24]

I have been using this project for a few weeks for a prototype of an app. I have seen that it might not scale too good for distributed systems, parallelisation, and large scale computation. Is this real or is there some way I could distribute HiGHS computation? I have only done an overview of the project so I don't know that much of how it works. Finally, is there any "in development" feature where i could start and contribute? preferably related to acceleration both CPU and GPU-based

[jajhall]

HiGHS is better for large scale computation than any other open-source optimization software.

What precisely do you mean by "distributed systems"?

Exploiting parallelisation for general problems in most optimization solvers is very hard, and even the commercial software seems limited to those from other fields. We are developing a GPU-accelerated LP solver, but the main challenge is the algorithmic features, rather than the GPU-acceleration. We are also adding a major CPU multi-threading feature to the MIP solver, but that's highly technical.

In general external contributions to solvers are not encouraged.

[ETROK24]

i mean integrating it, for instance in AWS services. So is there a small practical limit to the number of threads or kernels that can be used to accelerate a big problem or is it really scalable?

[jajhall]

No optimization solver is "really scalable" for general problems - even Gurobi

The only computational component in HiGHS that scales to anything like this 8-32 threads "plateau" is HiPO, our new interior point solver for LP.

It's the nature of large-scale sparse optimization problems and their solvers

[ETROK24]

Thankyou for the clarification. I was now reading a bit of the documentation, and I noticed I haven't specified enough.

I onlyuse the MIP solver. I just read docs/src/solvers.md and read "It is largely single-threaded, although implementing a multi-threaded tree search is work in progress.".

Is there anyway I could contribute to this tree search or at least start understanding how it is being done? I am a third-grade CS student and I mainly focus on parallelisation, although as you might imagine, my knowledge and capabilities might be short for the project

[jajhall]

Thanks for the offer, but it's taken almost 12 months of full-time work for someone whose doctorate was in computational MIP to get to grips with the solver and write a prototype multithreaded tree search.

[ETROK24]

Okey thankyou and good luck! I will keep up with news to know when is available