Originally posted by: Semen

I know there is phases in cp optimizer (IloSearchPhases) and with this tool, I can help to my solver to search the solution.  Have we are the same tools in cplex? or how can I make my solver faster? I have MIP. And I don't pretend on the best solution

or can I somehow point to initial solution?

You can provide some general guidance on the search process through the settings of a large number of parameters. These include which algorithm(s) to use, search strategy (DFS, BFS, best bound, ...), backtracking tolerance, and a ton more.

Using the programming APIs, you can also add callbacks that can apply heuristics to node problems, decide how to separate a node, to which open node CPLEX should branch next, etc.

My impression, based on significant experience with MIP solvers and much more limited experience with CP solvers, is that CP solvers tend to need more guidance in the search than do MIP solvers.

Originally posted by: Semen

thanks for your answer, I will try to use search strategy, thanks

Originally posted by: EdKlotz

I second Paul's comment that CP solvers tend to need more guidance than MIP solvers.   CP solvers do not solve a relaxation the way MIP solvers do.   This can be advantageous in certain situations, such as when solving LP relaxations is time consuming, or when the information provided by the LP relaxation solutions don't give good guidance for the MIP optimization.   It can also be disadvantageous if the LP relaxation info is useful.   But, in other case, the CP solve has less info, and may benefit from additional info provided by the user.

Regarding non default CPLEX parameters, first run CPLEX with defaults and see if it gives you adequate performance.   If you decide you do need to set some non default parameters, don't experiment randomly; they are too many combinations.   Look at the node log of the default run and find out which parts of the algorithm or not progressing fast enough.   Then choose parameters that will address that area.    For more details on how to do this, try the following link:

https://www.ibm.com/developerworks/community/blogs/jfp/entry/practical_guidelines_for_solving_difficult_mixed_integer_programs?lang=en

Originally posted by: Semen

thanks, i'll do it