[Coin-discuss] SYMPHONY speedup

[Tobias Achterberg]

Hi Jack,

for MILP, your question cannot be answered (LP is a different story). Parallel speedup and
performance difference accross different solvers depend on the model.

For example, there are models that can be solved by solver A in less than a second but
that take forever by another solver B, just because solver B is lacking the important
feature (like a cutting plane or a presolving reduction) to solve this particular model.
And there may be other models for which the situation is reversed. Finally, there are
models that can be solved in roughly the same time by all branch-and-cut MILP solvers.

Overall, my experience is that the algorithmic quality of a solver is usually much more
important than parallelization. This means, that often solvers like CBC, SCIP, Gurobi,
XPress, or CPLEX will be faster than Symphony, even if you use 1000 CPUs for Symphony.
But, as I said, some models do not require the sophisticated machinery that these solvers
provide and parallelize well. In this setting, Symphony can be really great if you have
access to a massively parallel computing environment.

So, the short answer is: you just need to test it for your models. You cannot say in advance.


Cheers,

Tobias

Jack Bryan wrote:
> Hi ,
> 
> I am a new user of SYMPHONY .
> 
> I need to use it to solve very large mixed integer linear programming 
> models.
> 
> So, I need to use PVM for SYMPHONY .
> 
> But, I do not know how much speed up I can get from the parallel SYMPHONY ?
> 
> For example, if it takes 100 seconds to solve a MILP model in CPLEX, how
> long to solve it in SYMPHONY ?
> 
> 
> thanks
> 
> Jack
> 
> April 21  2009
> 
> ------------------------------------------------------------------------
> Windows Live™ Hotmail®:…more than just e-mail. Check it out.
> <http://windowslive.com/online/hotmail?ocid=TXT_TAGLM_WL_HM_more_042009>
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-discuss