Threads in solving a MIP with Cplex

Solver related questions
Post Reply
Camilo
User
User
Posts: 1
Joined: 5 years ago

Threads in solving a MIP with Cplex

Post by Camilo »

Hi to everyone, i've been trying to improve the total time used in solving a MIP problem, and in my search i came to this link: https://support.gams.com/solver:multiple_cplex_threads
As far as i know the MIP problem is solved using a branch and bound algorithm that is parallelizable, and as the link states:

"The parallel option can be specified via the GAMS option threads:

option threads=8;"

So i tried using that option with the maximum number of threads that my PC has but i just don't see any improvement: the time goes from 9:30 or so to 9 minutes.

How can i check that actually the parallelization provides no improvement rather than i'm using it in the wrong way?

Thanks in advance and sorry for my english.
User avatar
bussieck
Moderator
Moderator
Posts: 1033
Joined: 7 years ago

Re: Threads in solving a MIP with Cplex

Post by bussieck »

Somewhere in the Cplex log you will find "Parallel mode: deterministic, using up to 8 threads." Also at the end you see a summary which you don't see if you run sequential:

Code: Select all

Root node processing (before b&c):
  Real time             =    0.25 sec. (48.12 ticks)
Parallel b&c, 8 threads:
  Real time             =    0.00 sec. (0.20 ticks)
  Sync time (average)   =    0.00 sec.
  Wait time (average)   =    0.00 sec.
                          ------------
Total (root+branch&cut) =    0.25 sec. (48.32 ticks)
Cplex and other parallel MIP solvers do a lot of things still in sequence. For example, solving the root node and processing node 0. Many models are nowadays solved in the node 0. Look at your Cplex log and see how much B&C it really does. Even if it does B&C nodes the improvements can be meager.

-Michael
Post Reply