# ConcurrentJobs

Filter Content By
Version
Languages

### ConcurrentJobs

Distributed concurrent optimizer job count
 Type: int Default value: 0 Minimum value: 0 Maximum value: MAXINT

Enables distributed concurrent optimization, which can be used to solve LP or MIP models on multiple machines. A value of n causes the solver to create n independent models, using different parameter settings for each. Each of these models is sent to a distributed worker for processing. Optimization terminates when the first solve completes. Use the ComputeServer parameter to indicate the name of the cluster where you would like your distributed concurrent job to run (or use WorkerPool if your client machine will act as manager and you just need a pool of workers).

By default, Gurobi chooses the parameter settings used for each independent solve automatically. You can create concurrent environments to choose your own parameter settings (refer to the concurrent optimization section for details). The intent of concurrent MIP solving is to introduce additional diversity into the MIP search. By bringing the resources of multiple machines to bear on a single model, this approach can sometimes solve models much faster than a single machine.

The distributed concurrent solver produces a slightly different log from the standard solver, and provides different callbacks as well. Please refer to the Distributed Algorithms section of the Gurobi Remote Services Reference Manual for additional details.

One important note about integer-valued parameters: while the maximum value that can be stored in a signed integer is , we use a MAXINT value of 2,000,000,000. Attempting to set an integer parameter to a value larger than this maximum will produce an error.

For examples of how to query or modify parameter values from our different APIs, refer to our Parameter Examples.