Is GAMS able to do parallel computing?

GAMS
1

Hi All,

I am doing stochastic programming research but I am relatively new to GAMS. One interesting I have found in GAMS is when you model a large number of Linear Programming (LP) problems and try to solve it, it would build each individually and solve sequentially. This process will consume a lot of time which are mainly the model building time. However, in other languages like C/C++ or Julia, one can easily pre-build models together in a parallel fashion and then execute solver. To the best of my knowledge, solving 10,000 LPs with 200+ variables and 300+ constraints in C with parallel computing in 4 processors and multiple workforces only takes about 40s, when the same problem in GAMS will be unsolvable based on my implementation (I tried to build and solve 600 LPs with 2 variables and 3 constraints in GAMS individually and the overall wall time is 700s). One relevant topic is as follows.
https://newforum.gams.com/t/individual-lp-v-s-integrated-lp-in-gams/2148/1
So, I wonder if anyone can show me how to do the same thing with C in GAMS of the parallel computing. Even a simple example will be highly appreciated. Thank you in advance!

2

Gabriel,

You may be interested interested in

a) The Grid and Multi-Threading Solve Facility for asynchronous submission and collection of models and their solutions
https://www.gams.com/latest/docs/UG_GridComputing.html


and

b) The Gather-Update-Solve-Scatter (GUSS) to allow collections of similar models (same model rim but different model data) to be described, instantiated, and solved efficiently.
https://www.gams.com/latest/docs/S_GUSS.html


Both chapters have a couple of useful links to the model library. One model that combines both approaches is GussGrid (https://www.gams.com/latest/gamslib_ml/libhtml/gamslib_gussgrid.html)

I hope this helps!

Best,
Fred

3

Thank you very much Fred! It helps a lot, especially the second post!

Cordially,
Gabriel