Documentation

Accessing your Cluster

Verify the Cluster Status

You can use grbcluster to display the status of your cluster, it will list of the nodes with information about the license and some important configuration properties. Please refer to the Remote Services Reference Manual for more details or run grbcluster --help:

$ grbcluster --server=40.121.109.191 --password=myAccessPassword nodes
ADDRESS        STATUS TYPE    LICENSE #Q #R JL IDLE %MEM  %CPU
40.121.109.191 ALIVE  COMPUTE VALID   0  0  2  1m   20.39 0.00

If you have started multiple nodes including compute servers and workers, you should see an entry for each node. Of course, you need to wait for the instances to start and join the cluster. Here is an example with a cluster composed of 2 compute servers and 2 distributed workers:

$ grbcluster --server=40.121.109.191 --password=myAccessPassword nodes
ADDRESS        STATUS TYPE    LICENSE #Q #R JL IDLE %MEM  %CPU
40.121.109.191 ALIVE  COMPUTE VALID   0  0  2  7m   20.38 0.00
40.76.55.208   ALIVE  WORKER  N/A     0  0  1  <1s  20.21 0.00
40.155.109.200 ALIVE  WORKER  N/A     0  0  1  2m   20.32 1.00
40.121.200.55  ALIVE  COMPUTE VALID   0  0  2  20m  20.25 0.00

Submitting a job

For a Gurobi Compute Server, you should configure a Gurobi Compute Server client to use the Gurobi Compute Server. This is explained in the section "Creating a Compute Server client license" in the Gurobi Optimizer Quick Start Guide. When running an Azure instance as a Compute Server, you simply need to set COMPUTESERVER to the address of one your Azure cluster node addresses. You also need to set the PASSWORD property to match the client password given during the node configuration.

When using gurobi_cl, you can also provide the address of one of the cluster nodes and the password:

    gurobi_cl --server=40.121.109.191 --password=myAccessPassword stein9.mps

Note that both COMPUTESERVER may define a list of nodes for better robustness so that the first reachable address will be used to contact the cluster. These values will not limit the dispatch or load balancing of the jobs within the cluster.