00:19:56	Siddhartha Ghosh:	You may write your questions for Irfan here
00:21:44	Benjamin A Cash:	You mentioned a GUI for bursting to the cloud - will that be available for scheduling jobs on the system as well?
00:24:49	Siddhartha Ghosh:	Hi Ben, the short answer is yes, from system side you will submit a batch job, perhaps in a queue (the details we are yet to learn)
00:26:08	Kevin Paul:	Users will be able to run Docker containers?
00:27:06	Benjamin A Cash:	It will be very interesting to see how that interacts with workflow engines like cylc/rocoto/ecflow/cime.
00:27:40	Brian Dobbins:	Ben, I’d be interested in chatting more about your cloud interests — GUI or no GUI, I’m thinking there may be some issues with running data-heavy models (like some CESM configs) in a ‘bursting’ mode, so I’d love to find out what your interests are, and how we can help.
00:27:49	Siddhartha Ghosh:	Hi Kevin, yes, as you know today we do run Docker container within Singularity and Charliecloud framework, so certainly yes, we will support Docker container
00:27:57	Mick Coady:	Kevin, yes the plan is for users to be able to run Docker containers
00:28:18	Kevin Paul:	Cool!  That will make things much easier for people.
00:28:40	Kevin Paul:	In less step in the process (e.g., no convert to Singularity step).
00:28:41	Siddhartha Ghosh:	Thank you Brian, that should be a good discussion topic in this forum.
00:29:05	Brian Dobbins:	@Kevin : You can also use Singularity now, on Cheyenne, to run Docker images.  :)
00:29:16	Brian Vanderwende:	Thanks for that Ben - we will be looking to test workflows important to users on Lab machines before anything is deployed locally. How workflow managers play with the GUI is something we will look to explore in those tests.
00:29:17	Brian Dobbins:	(No need to convert, either.)
00:29:49	Kevin Paul:	@Brian Dobbins:  You’ll have to show me how to do that.
00:29:59	Jim Edwards:	Is GPU direct storage available?
00:30:31	Ethan Gutmann:	How will cloud bursts interact with GLADE or other storage?
00:30:55	Supreeth Suresh:	@Brian Dobbins: I would be interested in learning that as well.
00:32:47	Brian Vanderwende:	Jim & Ethan - we will address those questions when Irfan is finished his slides.
00:34:58	Jim Edwards:	https://www.hdfgroup.org/wp-content/uploads/2020/10/GPU_Direct_IO_with_HDF5-_John_Ravi.pdf
00:35:08	David Gill:	You mentioned NDA on some issues. Again briefly what part was NDA
00:36:23	clyne:	Will the lustre FS be mounted on casper? Will glade be mounted on NWSC-3?
00:36:35	Brian Dobbins:	To add to what Irfan is saying, for CESM, we’re looking at adding cloud-based data stores for input data — so, *inputs* will be pulled from the local repo, vs. reaching all the way back to NCAR.  However, for writes to GLADE, this is problematic.  If this is a use case you’re interested in, let me know, I’d love to follow up with you (from the CESM angle, vs. Irfan’s system angle.)
00:47:43	Siddhartha Ghosh:	We are opening the floor now for questions and discussions
00:48:02	Ethan Gutmann:	The cray environment can be tricky, can / will we get access to a test system for compilation ahead of time.
00:50:44	Benjamin A Cash:	When do you anticipate the call for ASD proposals to go out?
00:51:09	Brian Vanderwende:	We will have an announcement about that very shortly Ben, but short answer is in the Fall 21.
00:51:19	Benjamin A Cash:	+1
00:52:55	David Gill:	There has been discussion at a number of levels at NCAR on the topic of retraining staff for utilizing GPUs. Do you see that fitting under the user applications?
00:53:39	Jeremy Sauer:	Is there any notion of how allocations and charging will work for the GPUs?  Will GPU nodes be a separate queue from CPU-nodes?
00:53:55	Brian Vanderwende:	Absolutely Dave. We are doing internal training on GPUs right now (today we have one in fact!)… once we have finished that, we will be focusing on offering a wide range of user training.
00:54:06	Mick Coady:	Dave - I’ll address your question when Sidd is done with the current question.
00:55:13	Benjamin A Cash:	Yes, thanks.
00:56:14	David Gill:	@mick Thanks
00:56:57	Jim Edwards:	Can you talk more about how glade and the new filesystem will interact - what happens to glade when Cheyenne is decommissioned?
01:01:34	Jim Edwards:	Thank you
01:02:18	clyne:	Yes, please make the topic of storage an agenda item!
01:02:19	Gary Strand:	Yes, let’s discuss the storage stuff.
01:02:43	Siddhartha Ghosh:	Ok, sure, we will do both @gary and @clyne
01:02:48	Gary Strand:	Thanks/.
01:03:15	Charles Jackson:	Will there be topics for these meetings?
01:04:50	Jeremy Sauer:	328 A100 GPUs to begin.  How much room for more assuming a need for expanding GPU partition within NWSC-3 lifetime?  A doubling?
01:06:18	David Gill:	Of the 70+ folks on the call, approximately how many are from the university community. Basically are they well represented.
01:06:28	Jeremy Sauer:	@Sidd great information thank you!
01:07:05	Skyllingstad, Eric:	What is the current GPU utilization?
01:10:16	David Gill:	@sidd That sounds like a well rounded group
01:10:39	Jeremy Sauer:	@Skyllingstad I can’t say a pure utilization number, but I can say there is a significant increase in utilization over the last 6-8 months.  Definitely at times there are one queue waits due to higher demand for the very limited 64 v100 GPUs of Casper.
01:11:34	Jeremy Sauer:	@Brian well said.
01:12:43	Jeremy Sauer:	Thank you
01:12:52	Hunter Brown:	Thank you for putting this together!
01:12:54	Melanie How:	Thank you