Re: [Help-glpk] GLPK re-endrant

[Rios, Joseph L. (ARC-AFO)]

Hi Larry,
 
> I apologize if I'm not up to speed on multi-threaded and re-entrant coding.
> What is the purpose you are trying to accomplish with your patch?  Is it just
> the ability to multithread?

Most definitely.

> Or is there a possible benchmark improvement to
> GLPK?

I don't think the parts in my patch have anything to do with the performance
(speed) of GLPK, nor were they intended to.  And I personally don't have an
interest in benchmarking GLPK.

> Just wondering what is lacking to make it on par with CPELX or whatnot.

I'm sure Andrew would have a better idea of this than I would.

I have profiled the code a couple of times, and if I'm not mistaken a good
deal of time is spent in LU-factorization stuff.  It looks like it was coded
from scratch for GLPK, which is cool.  But I think that using a library
(open source or not) optimized to do this work would pay some speed
dividends.

Can anyone comment on the appropriateness of that assessment?

If you are super-interested in this topic, I'd recommend profiling GLPK with
a bunch of different problems and see what functions are gobbling up the
processing time.

Joey