This shows you the differences between two versions of the page.
|Both sides previous revision Previous revision Next revision||Previous revision|
solver:is_it_possible_that_a_concurrent_execution_takes_longer_than_a_non-concurrent_execution_for_the_same_model [2007/11/20 19:17]
|Line 1:||Line 1:|
|-||====== Is it possible that a concurrent execution takes longer than a non-concurrent execution for the same model? ======|
|-||- Solving an LP with concurrent is always as fast as solving sequentially with one of the methods run in concurrent mode, but the ''resusd'' can be much larger for the concurrent since we report real time in the concurrent and CPU time in the sequential run. Depending on the load of the machine this can be quiet different.|
|-||- If we talk about parallel MIP (parallel solve of node LPs), then the parallel run can be slower than the sequential, since the parallel can go a different route in the algorithm leading to poorer performance even if the operation are carried out in parallel. The 'deterministic' mode is not helping here. It just guarantees that we get the same performance with multiple runs. It can still make a poor decision at the beginning of the B&C tree resulting in poorer performance. We have seen few examples where this happens, but overall parallel pays off nicely.|