Open kwmsmith opened 10 years ago
This is a duplicate of issue #169, I'll close #169 but ask my question here again.
What does @enthought have in house that we can use? Can TACC run ipclusters?
We can do somewhat larger scaling tests with in-house workstations. This will allow us to test usecases where the client and engines are not on the same machine.
It will likely be easier to implement the MPI-only communication layer and use that for testing on TACC rather than go through the trouble of getting an ipcluster up and running. TACC does have ipython, but it's an old version (0.13.1), and I'm pessimistic about how it would perform vs. an MPI-only version.
We did do some testing of separate client and engines. See #375.
Components in place to run MPI-only contexts on TACC which remains to be done. Will do if time permits. This milestone is otherwise addressed by allowing separate client / engines.
Conclusion for 0.6- some has been done, we should definitely do more. Moving to 0.7.
Objective: To determine the steps necessary and usability constraints when deploying ODIN on a distributed cluster.
Relevance: Ensuring that ODIN is easy to deploy on distributed clusters and supercomputers ensures that end users can access ODIN’s functionality on distributed computing resources.