Thank you.

Can you please add me? Thanks.

Throw on this the fact that some applications are license restricted, and again we don’t want a single user to take all the licenses.
All this means that simple metrics like wait times are useless. And you want to find the right mix of utilization and availability.

And then you have the power users that submit jobs with thousands of cores, and sometimes a lot of those. So they don’t expect all of those jobs to finish at once. Then you add to this that you have limits on users to make sure a single user on Sunday evening can’t take over the system for the week.

This is one of those things I struggle with a lot. Considering that we have a large user base with vastly different demands finding good metrics is so hard. Some users only need a few cores for an hour, others submit thousands of small jobs and don’t expect them to come back all at once.

Yeah, it’s funny when you talk to cloud providers about doing HPC and they are asking for forecasts. Why do you need a forecast when cloud is supposed to be unlimited?

In industry it is hard to find the funding to port to different architectures all the time.

That is so ridiculous. Sorry that you have to deal with that. I thought our industry was better than this.

That is so true, especially for the national labs. We always looked at them to bring in the experimental stuff, and then a year later when a lot of the bugs and weird behavior was worked out we would get the smaller versions of those systems.

Yeah, hopefully next year I will be able to go.

Yeah, it’s weird not going this year.

Thank you.

Please add me as well. Thanks.

I would expect Azure again.