[Cbc] setting setUseElapsedTime caused RINS to immediately time out

[John Forrest]

Fei,

Fixed in trunk and stable/2.9.

If you need it in 2.8, it is only a two line change - look at the 
changeset for 2.9

John Forrest
On 28/01/15 16:30, fpeng wrote:
> Hello Cbc community,
>
> I'm seeing an issue when using wall time instead of cputime in CBC: 
> after calling setUseElapsedTime(true) (which should make cbc use wall 
> time instead of the default cpu time), all RINS runs during cbc 
> branchandbound will time out almost immediately, something like:
>
> Cbc0028I Starting sub-tree for CbcHeuristicRINS - maximum nodes 200
> Cbc0020I Exiting on maximum time
> Cbc0005I Partial search - best objective 1e+50 (best possible 
> -391348.59), took 0 iterations and 0 nodes (1422388648.51 seconds)
> Cbc0035I Maximum depth 0, 0 variables fixed on reduced cost
> Cbc0029I Ending sub-tree for CbcHeuristicRINS
> Cbc0045I Heuristic RINS took 0.139259 seconds (no good)
>
> although the rest of the cbc run seems normal and uses the correct 
> wall time to terminate without a problem. I used the default RINS and 
> didn't change any settings. I believe RINS smallbranchandbound is 
> using the main branchandbound routine, so it's perplexing to see the 
> time handled incorrectly here.
>
> Some background: the reason I wanted to use wall time is that cpu time 
> doesn't translate to wall time very well in parallel: if I have 
> multiple cores, setting setMaximumSeconds(tLim*numCores) doesn't 
> always terminate cbc after tLim seconds, and sometimes I get vastly 
> different termination times. If anyone knows what's happening here 
> please let me know.
>
> Thanks!
> Fei
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc
>
>