Documentation

Referring to Compute Servers

One important difference between version 8.0 and previous versions relates to clustering. Previous releases supported queueing and load balancing among multiple servers, but this clustering was handled on the client side. Your client program would explicitly list the Compute Server nodes that your job could run on, and the client would negotiate with those servers to determine where and when it ran. In the new Compute Server, you form explicit clusters of nodes, and your client simply needs to refer to any member of the cluster in order to run a job on any node in the cluster.

Another important difference is in how you to refer to Compute Server nodes. While it has always been the case that you need to provide both the name of the Compute Server node and the port number that Remote Services is listening on, those were separated in previous releases. For example, you might say:

> gurobi_cl --server=server1 --port=61000

In version 8.0, communication happens through a REST API, so you now use standard HTTP syntax to refer to servers:

> gurobi_cl --server=server1:61000

...or...

> gurobi_cl --server=http://server1:61000