Consider downsizing the default pool machine type

[ml8]

Right now we have an n1-standard-2 running as the login node. We are probably fine with n1-standard-1.

This is a change to the machine type for the default pool for the cluster. The user pool is n1-standard-2 to try to keep slots for new arrivals (users get half a core).

[ml8]

It looks like there's about half a core underutilized per node, but memory is at 90%+ utilization. Might consider a custom machine type or just giving a guarantee of a quarter core with a cap of a half core (current cap is 0.25).

[ml8]

I made a dashboard.

It looks like both CPU utilization and memory utilization are less than 25%. 95th looks heavier, but still not above 50%.

The current config is:

• memory: 1G limit, 1G guarantee.
• CPU: 0.25cpu limit, 0.25cpu guarantee.

Given that, I think it might be reasonable to downsize the machine type to e2-standard-2 and have the following:

• memory: 1G limit, 250M guarantee.
• CPU: 0.25cpu limit, 0.10cpu guarantee.

Of course, this would be an experiment, and the real test would be during a period of high load (afternoon on a MWF or evening when assignments are coming due).