National Energy Research Scientific Computing Center (NERSC) NUG Monthly Webinars, 19 Nov 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: NUG meeting 19 Nov 2020

Description

Recording of NUG Monthly Meeting for November 2020

A

That's it especially because we've got a a smaller number of us. um Yeah please participate. Let's make this uh yeah very much a discussion rather than a one-way thing.

A

So our um kind of regular format for these, as we go through kind of a win of the month, which is you know, um celebrate people's achievements um today.

A

I learned, which can be either either a yellow, celebrate our failures or you know, um talk about things that we've discovered and- and this can include things that we've uh come across and- and you know, read and learned, as well as uh yeah things that we've tried that have proved more challenging than we expected and that you know good good tips for the rest of our user community.

A

um A few announcements and calls for participation, uh then we'll spend sort of you know 15 minutes going through our our topic of the day and for this month it's currently nearing the end of the sc20 conference and so we'll make that our topic we'll talk about showcase, a little bit of nisk's involvement in that and then a bit of preparation for upcoming meetings and a quick look at our numbers for the last month.

A

So the idea of the win of the month segment is this is an opportunity to show off an achievement or shout out someone else's achievement that you know about such as you know, having a paper accepted accepted. um Something is your seemingly, um you know small scale as solving a bug that you know you'd been working out for a while that was proving challenging, and you know, especially if there was some uh yeah good tips and lessons for the rest of us there, um scientific achievements.

A

um This is also a good uh opportunity, an avenue for um science highlights or um yeah science uh uh high impact scientific achievement awards, or also innovative in innovative use of high performance computing awards which uh nursk awards each year.

A

uh Does anybody have anything that they'd like to either nominate self nominate or shout out somebody else for.

B

Well, this is koichi from pinal uh hi. This.

C

B

Yes, yeah, we had a paper uh accepted actually already published uh earlier this month or maybe last month, but I I forgot to prepare the information on hand, but maybe uh next month's uh maybe show a figure or something, but that's totally, uh based on the simulations. We run on nas machine, let's uh climate model simulations, but with relatively new uh model or more like experimental, so such that it's global model, but we can refine the grid spacing or special resolutions over a certain region of the world.

B

So for this paper we use three simulations only two three simulations. For example: it's globally uniform uh 200 kilometers with spacing outside the world, but only only over the us. We refine the grid spacing to, for example, 25 kilometer uh with spacing, which is getting close to the scales for cluster resolution. We can resolve those intense large uh convective stones and particularly important is the we were studying this particular kind of stone which called mesoscale convective systems.

B

That's really uh lead to many flooding event over the us about 60 percent or 50 percent of the uh annual precipitation. Precipitations are coming from this storm so using those simulations around mask, and then we also developed a algorithm again around on the mask to automatically detect and track those mesoscale convex system, both in observations and simulation. So this paper is actually technical paper that we describe the technique. We use to track those stones in the simulations and compare to observations and how the model is doing.

B

uh So I believe this is gonna, be a maybe you know really useful one to the community and uh hopefully it's gonna be uh cited and we do it. We did acknowledge and ask on this paper so yeah. That's.

A

B

Yeah, thank you.

A

So so yeah 25 um kilometers would be yeah. So I guess that's. The um yeah, like large storms does that capture hurricanes.

B

They start to do that.

C

B

Some you know still, there is some differences in the maximum wind speed, for example, but yeah, because it start to really deserve hurricanes. uh Actually, uh the next phase of you know ipcc's climate warming report, uh there's a bunch of us around the world contributing climate simulations to for their scientific basis, and then this time. So this phase is gonna, be a sixth uh climate report.

B

We are international committees, including the particular model archive that is specialized in higher resolution global models and that grid spacing is actually around this 25 to 70, kilometer, yep and then and uh people are already started.

B

uh You know looking at some of those model, outputs, that's already available and then uh try to study how the hurricane is gonna, be changed in in the warming future of 50 years or 100 years later, based on those a bunch of you know, higher resolution global simulations- and I am actually running these simulations and asked for steve and as us to actually uh to better uh manage workflow and make it reproducible and sustainable and sometime maybe next year. I wish I can demonstrate how useful those model outputs are in this month's remaining.

A

That's right: yeah. We were talking about um building the model in a in a container, perhaps to keep it yes, um sort of stable version and so on over over a long period of time that that should be quite an interesting project. Yeah.

B

Yeah, I think I'm really excited it's uh and then because those simulations are kind of scaled in 100 years it takes a real. You know really couple of years to finish so yeah really looking for this quantitation of our model, executable.

A

Yeah- and this is um so- I think we were talking about the csm model for that which is which is a fairly complex model, um which, which is what makes it kind of an interesting challenge compared to um containerizing a uh you know, a lot of our other applications um so.

B

A

um The mesh refinement system was that um built into csm as well. Was it a standalone, separate model? I'm.

B

Sorry, could you say it again, the question the.

A

um The mesh refinement system that the.

B

A

Was about? Was that also built on on cesm.

B

Not yet actually the work is ongoing now, uh so they are really working on bringing this particular model component as uh official part of the released version of csm, maybe next version. So we are kind of a guinea pig running those models in the existing csm. What kind of uh you know complications or problems we might encounter, and we did a couple of uh issues in running this model, even with higher resolution down to four kilometer.

B

It is at this resolution. I am seeing some memory scaling uh problems so uh right now I cannot run continuously this four kilonewton high resolution uh simulations on nurse corey or other systems at the national center for atmospheric research.

B

But, interestingly, before we had this major updates several years ago uh before s islam system, we are using something different right. uh Yep. At that time, yeah I forgot the main vendor every.

A

Ap run maybe.

B

Yeah and at that time this mpi available npr library available at some time. We did not have this memory problem, but once we change, then we suddenly start to get this more memory problem and it's coming, maybe from some npi some routine. It just gets uh frozen, exchanging some data and I just didn't- have a chance to open the ticket for this, but uh maybe in near future. I might try to you know, work closely with you guys to solve this problem.

B

Then that gives us really exciting research venue, you know going down to four kilometers, really makes model much more reliable to predict those uh storms.

A

Yeah sounds good um yeah. I reckon for for debugging that we'll um take that into a ticket that should be a yeah one, that we uh dive into and um yeah a great way and and four kilometers you'd be able to um resolve individual thunderstorms. I guess yes, it's given that yes, yeah cool um good. Thank you yeah thanks veggie, um so I had reports from um another of our users.

A

uh Dilip a siguri, I'm not entirely sure if I'm pronouncing his surname correctly, but I had a paper just recently published in journal of physics, physical chemistry, letters that is a fairly significant finding around uh protein folding and showing the influence that temperature has on protein folding. So um there was a press release around that, but a lot of that work was done using nest systems as well, so shout out to dillup on that. We've got a link to the press release in the article here as well.

A

um Does anyone else have a story to tell us.

A

About, if not we'll go on to the next thing in our um agenda, which is sort of the other side of it um today, I learned what surprised you that it might benefit other users to hear about you know and and uh along the way might help nurse to you know improve our documentation, for instance, or you know, make make tweaks that can make our system easier to use.

A

I have to say this week or this last couple of weeks. I have not spent a lot of time uh bug hunting, for instance, because I've been fairly occupied with the super computing conference which we'll talk about in a little bit, but there was definitely a a lot of interesting things to learn there immediately. I don't know how public there there's a youtube channel um that I suspect a lot of uh talks will appear on.

B

uh So again, this is koichi again yep. I'm kind keep keep uh uh bothering but uh yeah.

C

B

So like I, we just briefly discussed. We are starting to work on using you know, shifter for our climate model. So first I learned uh this month. I seriously more seriously. You know uh took some tutorial about what is really image and containers and then when went through shifters documentation, which is very nice because they have several examples that uh but uh again it's just more like uh suggestions or maybe in the future, uh including my you know. Maybe your potential work, but uh just the application for the main application is probably the python.

B

I guess for the shifter is that correct.

A

um Shifter is is particularly good for python things, um and particularly at scale because of uh some of the, I guess um gotchas with using python and um dynamic.

C

A

Yeah, it's dynamic loading at scale that and yeah shift solves a lot of those.

B

Okay, yeah, I'm thinking to use, maybe in the future. You know when I need to scale my past application and particularly if we have to share that you know, when we publish a paper, we have to account for some reproducibility issues, so uh this is a really good, uh really good. I think it tunnels out how to share the methodology I used and so that other people can reproduce the results and uh for documentation itself.

B

Sometimes I find some part of the documentation was rather difficult to understand, so I quite uh you know needed to quite a few google search and then go through some other technical. You know articles, so um I wonder if there is any places I can put questions or comments for particular part of documentation like this paragraph, it's not too clear, or does this mean this or maybe you can edit this? Yes,.

A

um So so, actually the nurse docs pages are a public git lab or you know, come from a public gitlab repo. um So you can um in fact you can make merge requests against it. So you know you can uh add to its improvements as well.

B

And I do have a link here.

A

B

So I just need to download this. uh The git.

B

Just push the change or suggest.

B

A

Yeah, basically, you can um make a make a mode request or uh open an issue. Even so up here. There's a link to the gitlab.

B

A

Site was going to ask when you were talking about doing tutorials, um which tutorials were you following and I guess did you find them helpful and if so can can you point us at them? It might be good.

B

For other users too, I think I just follow the link from the mask documentation. They recommend tutorials of a docker somewhere.

B

So I followed that, and the docker website has a bunch of a nice tutorials, including huge youtube videos, and the other is. This is my surprise, but there's already uh you know docker image of the csnt model in a simplified configuration.

B

So I I kind of use that uh that is publicly available from the national center for tech research that they use, that the docker image of these components, kind of models for for tutorial for the students, and that's also helped me to really get the you know, understanding of doctor image and container, even though I'm still really debating how people use the term containers and image.

B

Container is just one, uh you know realization, or instance, of an image, but that's not how people generally use these terms does not really sound like it do. Do you have any do you know.

A

So somebody else on the call might have better experience than than me at this, but I've always taken the um understanding that the image is the thing that's on you know, on the on the registry or on docker hub or something like that, and the container is the running instance of it.

A

So, okay, so shifter runs the container okay, it's possible that somebody else here can correct me on that. Okay,.

B

That makes sense.

A

Okay, so that's a that's a good tip that incar has a good tutorial.

B

Yeah they did. I didn't know that, and I was really surprised about this docker image of this complex model.

A

Yeah, that sounds like a good place to start.

A

um Cool thanks scratchy um time for our next um item, which is announcements and cfps. So there's a few things that we know of that are or that are um published in the um nurse weekly email. That I'll just uh shout out to one is. You should see in your inbox this week sometime an invitation from a company called nbri on behalf of desk for your feedback. This is our annual nurse user survey and um it is really important for us both in terms of our reporting and keeping.

A

uh uh What do you call it discovering how we can improve our services to users? So uh please follow that. When you see the link, um you probably noticed there was a maintenance. Yesterday, cory was out because it's back again now, the maintenance was was, I think, mostly fairly minor from a user facing point of view.

A

The pe defaults are still the same as what they were, but if you were using the old 20.03 you'll find that that's no longer there, but you can use the new pe uh cdt, sorry creepy um 20.10, but neither of those are defaults.

A

You need to you know, do a module load to access either to access that um we have some upcoming training events at the beginning of december, so between december 8 and 10, there are training events for nvidia's hpc software development kit and also for the total view, hpc debugging tool um so yeah that that will probably be very useful for anybody who's trying to do debugging at scale.

A

Big one coming up in december, there is the final power upgrade in preparation for installing installing polemata and that's going to happen in the week of uh december 15 to 20., so in particular we're expecting power to be out between tuesday, the 15th and friday, the 18th of december, and so that will affect all of nurse, not just corey everything file systems, hpss networks- um it will have- will have no power in the building, so keep watching the weekly emails for updates on that, um but yeah be be prepared for a outage. During that time,.

A

um We're coming close to the end of 2020 and there are a couple of changes happening in 2021. A particularly important one is changing the way that the premium queue is used uh lisa. Do you want to expand on this.

C

um Sure I can um so we we have a a staff committee that needs to assess sort of the the queue and whether it's meeting users needs, um and one of the things we wanted to do was to reassess how premium was used so that it's really just used purely for a scientific type emergency um or you have something like an unexpected scientific event like a supernova or maybe something happens in your detector, and you really need to to push your jobs through.

C

That would be something that you would you would try and use premium for you know you have a review and you need. You need the results from that review right away, something like that.

C

We did find that some repos were using tended to use premium at the end of the year, just to kind of use up their hours, and we don't think that's really how we think it should be used. It should be available to all users for urgent scientific scientific needs.

C

So, starting next year with the new ay year, we're going to change the way that projects can use premium.

C

So normally the charge factor is times two and so folks will continue to be able to use premium at a charge factor of times two until they spent 20 of their total allocation uh on premium jobs and then, after that, the charge will increase to times four um and we we may reserve the right to change this um increase in response to how premiums being used, and one of the other features that we're adding for this is that pis will be able to go into iris and um select whether or not users in their projects will have access to premium, because sometimes people you know accidentally use premium or use more than they intend.

C

So now. This will be something that the pis can toggle on and off and we'll have instructions on the web page on how to do this by default, all the users won't have access to premium so you'll need to if you're api you'll need to go in and turn on. If you want to give someone access, you'll need to go in there and click the couple buttons.

A

Cool thanks lisa that uh I think will be very handy for a lot of pis in in managing their allocation.

A

um We also have a new jupiter kernel, a new julia kernel for jupiter and johannes. I think you know details of this one.

D

Yeah, so we've added support for julia to our jupiter hub.

D

So if you go to jupiter.net.gov now, you will see the julia kernel appear in the list of possible kernels to choose from I've sort of highlighted with the errors here. So these will be officially supported kernels by us and they will live alongside your.

D

If you have any of your own local kernels installed, um they will not live alongside those. um The um quote-unquote official kernels have the additional advantage that they will give access to the julia packages that we support at nurse alongside your packages that you have installed locally this. This is a new feature, so I am looking for people to kick the tires.

D

So, for example, some some bits in corey gpu are still experimental, such as mpi, and so, if you have a chance to play with this and find out anything that doesn't work or if you have any requests, please let me know- and I'd be really happy to hear about what your perspective is. Thank you very much.

D

Oh uh one, one uh final thing: uh um we sometimes see uh the kernel takes a little bit of time to start up, um so I'm just going to uh encourage people to be a little patient with that. This is the standard, julia problem, uh sometimes uh the kernel it has to pre-compile some packages, and so, if it looks like it's stuck during the connecting phase, um give it a moment that might just be it pre-compiling stuff all right. So that's it.

A

Cool thanks, johan, so and and that kernel can see packages in julia. So if you do package uh package.ad yeah, um it will appear in that kernel as well.

D

Yeah so the way we set it up is um we have um a um sort of a commonly supported packages list that we've installed for corey, and that contains things like mpi that need to be configured for for our systems um and then, when a user uh goes and says package add I don't know, let's say distributions or something that will then land in their um dot. Julia folder right.

D

So so julia sees the union of our package list and whatever the user has uh installed uh for their local system like for in their local environment.

A

Sounds good. Thank you.

A

A

So that's all of the announcements that I knew about are there any others that people would like to make that I didn't know.

A

A

So, if not, then so for our topic of the day this week, so this this week and last week have been the um supercomputing conference sc20 so also also known as the international conference for high performance computing, networking, storage and analysis. It's. uh um What do you call? It is domain grows every year. It seems um but yeah, that's uh it's quite a major conference for hpc and nurse has been quite involved and I suspect a lot of nurse users have also been involved. um Yeah. Most of our users, of course, will be working on.

A

You know. Science, that's in you know a specific science domain, not you know they're. You know people are using computational science to achieve.

A

um You know research and outcomes in other fields, whereas this conference focuses on the computational aspects of the computational science itself, so I thought it would be good to just do a quick run-through of um some of the things that happen in the conference and what nurse's involvement is it in it is and nurse users so um also very interested to hear um you know from our users if you've been involved in anything that you've uh done so by way of overview.

A

This is sort of what the web page looks like google search will find it pretty quickly.

A

This year, being you know, covert affected where it's a virtual conference and it's spread over two weeks instead of the usual one week. So and it's it's got quite a wide. um You know range of materials, tutorials workshops, um events for students, social things and uh you know, papers and and panels.

A

So one um fairly big aspect of it is tutorials um in the first week- and these are uh quite you know, deep dive intensive sessions that are, you know, a full either a half day or a full day, um and nurse staff and nurse users were involved in a number of these, so uh performance tuning with the roofline model um visualization, and it's in situ analysis, parallel, io, openmp, deep learning, upc plus plus containers.

A

So I think that was all of the tutorials that I knew that um no skin excuses were presenting. So that's quite a few there's. Also. A lot of workshops happened in the first week of the conference this this year and nurse staff and users were involved in quite a lot of those as well, and I suspect that I haven't even caught the ball here. So there's the women in hpc workshop um this year there was the first international workshop on quantum computing software and some nurse as well as other obl.

A

People were involved in sort of organizing and presenting at that there was the uh deep learning on supercomputers workshop um husk is the international workshop on hpc user support tools? So that's one that nurse is sort of directly involved in there was the essay workshop on best practices for hpc training and education.

A

There was a workshop on workflows for large scale science, a workshop on containers for hpc.

A

There was the parallel data systems workshop um workshop on performance, modeling benchmarking and simulation workshop on accelerator programming, uh interactive high performance computing such as jupiter, um machine learning and experiment in the loop computing, which is um computing or supercomputing, for you know large scientific facilities, things like accelerators and and telescopes.

A

So I guess the um the message here is that you know there's a lot going on, and particularly your nurse good nurse users uh involved in a whole lot of different aspects of or usage uses of hpc, as well as development of it out of interest. Was anybody yet something that I missed here.

A

So as well as the workshops, there was uh the main paper session and that's been happening this week, kind of in the in the technical program. um I sort of captured a few papers that nurse staff and nurse users were involved in from the main paper section. I haven't captured the ones from the workshops, I suspect, there's even more of those, but so there was uh some that nurse staff were involved in.

A

uh Some of them are kind of domain science, such as the ones you can see on the on the right hand, side of the page here, um and some of them are computational science and hpc management, such as the ones on the left.

A

Some nurse staff were involved in a gordon bell effort, that is a finalist, and the award ceremony actually happens immediately after this meeting, starting at 12 pacific time and we'll find out who won that one.

A

Nursed users presented even more papers, so there was a you know: a number of papers that I saw at least bits of that used nurse systems.

A

Across sort of a mixture of um so yeah computing and the domain sciences, I spotted nurse staff names on at least three posters, and I suspect that there's more from nurse users that I didn't see.

A

There's a fairly big students at supercomputing program and one aspect of that that so actually nursk was involved in a number of aspects of the students program. This is kind of important for bringing up. You know, uh fostering new talent in hpc and in computational science and and one aspect that.

A

Is is due to my heart that I was involved in? Was um we hold a well? There is a student cluster competition that this year was entirely virtual, so everybody was building clusters in azure and running a set of benchmarks and applications, and so in fact, one of the applications that the students had to build- and you know, run some tasks for uh was in fact cesm, um so yeah reproducibility challenges and other other applications and activities.

A

So this is a a dashboard that kind of helped to keep things exciting when you're in the yeah, the the somewhat more distant um virtual environment compared to physical. um But you know being able to see the teams. You know burning their cloud dollars and using cpu time and getting results in um you know was, was quite engaging.

B

Oh excuse me steve yeah. What did you say? What was the application that used for this student competition.

A

So, each year, there's uh two or three applications, um there's generally there's two applications that are announced in advance. One application: that's only announced at the competition, so it's a mystery application and one reproducibility challenge.

A

So our two applications um that were announced in advance this year were gromex and cesm, so the students had to build and run a number of um tasks on each of those. The mystery application was one called minivite, which is a believe. It's a graph analytics program application. I haven't sort of looked too closely at it, and the reproducibility challenge was, um from a paper from last year, called mem xct, which is a.

A

Memory, oriented or memory optimized.

A

X-Ray tomography, reconstruction.

B

A

um They used some fairly clever techniques to uh run at very high scale, so the students attempted to reproduce some of the experiments that that paper published.

B

Okay, good interesting, thank you.

A

Yeah, so actually one of the things that has been a bit of a topic of interest for a little while now is this whole idea of reproducibility and sc papers are now they were encouraged. I think they might now be required to submit with the paper um artifacts to enable reproducing the research.

A

So yeah that's a kind of an increasingly important aspect of computational science. um There was a a bunch of also um birds of a feather session panels.

A

um Nurse was also involved in a number of the committees, helping to organize and run the conference and a number of um nurse staff, especially, I think some of our networking and security people were um instrumental in cy-net, which is the supercomputing network, which is a a fairly major effort to you know enable kind of yeah high-speed networking to support you know, especially the exhibit floor as well as all of the connections that are being used by you know, 10 or 15 000 people at the conference.

A

So I guess my key sort of take home here is um that nursk is really involved in the development of hpc as a field. So I you know for, for our users, nurse is kind of a facility that provides resources, and you know, hopefully helpful interactions with with hpc support and so on.

A

You know to enable doe researchers to uh you know to to make great discoveries, and you know, support the work that you all are doing, um but yeah. I also wanted to kind of uh show off a little bit.

A

That nurse is also really actively involved in the development of hpc of computational techniques, of how to better run a high performance computing system, and you know the the future directions for it to go, um and- and actually our users are a pretty important part of this, because you know the the needs and interesting ideas that you all come up with um kind of you know help us help, you motivate our um your directions that we try to push hpc in and then in turning kind of a virtuous cycle.

A

The outcomes that nurse users achieve uh yeah come back around to the funding and support from government that you know enables nurse can and similar facilities to keep on going. So, thank you all.

A

um Which kind of leads into, or actually maybe before before, leading into that? um uh So I I see in the participant list there's a number of nurse people here who were probably at the conference. I don't know if any of our users were there, but does anybody want to shout out any other particular sessions or involvements that you saw or were involved.

A

A

A

Silence there, that's that's all good, um so this kind of spills into uh our next agenda thing, which we've sort of skipped through pretty quickly in the last couple of these meetings, because we've uh you had a somewhat involved topics of the day, which is requests and suggestions for upcoming topics, and so in a way.

A

So what we just did here was a showcase of yeah, something that nursk has done and we're pretty interested actually in hearing also from our users about your work and how you're using nurse facilities- and you know what you're finding works for you what's challenging and how you're overcoming those challenges and just you know a bit of a showcase, the research that you're doing and the outcomes that you're getting so yeah we're very interested in topics for future meetings and I'd like to really encourage um our users to you, know, put a hand up and um yeah present, maybe a a five minute lightning talk on.

A

You know what you're doing and uh yeah show off uh what you're achieving other than that.

A

um So we have a few ideas for topics coming up, such as more about slurm uh in a in a conversation the other week. Somebody was talking about the hpss move and you know that's probably another quite interesting. um You know story to tell, but I'm interested to hear if people here have any other topics that they'd like to nominate or request as a topic of the day for those meetings.

B

Steve, I think I have one a couple of uh ideas or questions I want to discuss in this mastery meeting.

B

One is the uh what is the best practice or a good template for users to open a ticket like when a user open ticket? What kind of information should be or is, is better to be there in the first time so that you know not skeptics can more easily understand situation, yeah and then so that's you know so how to describe the problem and also um I sometimes get wonder.

B

How can I tell how much I know about the program to the help desk as well or more in general, I'd like to know if there's any general idea of level of the user's expertise in hpc. Like you know, we have many users, many different backgrounds, I believe at the mosque.

B

Are there any general idea of from ask side to some assumptions assumption about? Maybe users know this much or maybe they should know some. You know, of course, I assume most of the users know basics in unix, for example linux, but uh some uh students may be first time to to even use hpc. So some people might abusing lowly knows, for example, uh also one time I feel sometimes I feel this when I read in documentation.

B

uh Documentation explains something well above my knowledge base right. So that's one another aspects of uh so so. First, we have to know, probably what is that you know average knowledge of the users but which also difficult to quantify. So really. I don't have a good idea to measure this, but at least from uh coming back to the first point, it's I think nice to know just to share with the users what information would be helpful if we, when we open the ticket, and so I do.

C

B

A

B

That with other users and not ask stuff.

A

Yeah, I think that would be a a good topic, actually um might be a good one to target in the early part of the year um when we, when we have you, know a whole lot of new users coming on.

A

I think something that we've seen over the years is that the um the nurse user base has expanded from a you know, a smaller number of kind of your relatively highly experienced people in hpc to a much wider range of people who have come from very different backgrounds and may not have had as much experience in hpc so yeah, something that you may have already seen that we kind of hope will make this a little bit easier.

A

Is the new service portal um at the moment, I believe it's still in in beta, but if you add sp to the end of the web address, you can see that.

B

Okay, yeah, I I think I tried I'm opening ticket for the new uh interface these days. I I like it quite a lot. Yes, it's really nice good job good to hear.

A

B

But just one more comment about the you know: the the users. uh I noted many students using in the atlas in my field, because models are so nicely configured for autobox.

B

uh You know simulations yep. You can run pretty much simulation, particularly in a computer system that is supported by csm, for example, without knowing much about hpcc or hpc system. So once they start doing something else that there's some you know learning curve and then some potential to make some problems.

B

So yeah yeah uh these some. You know, schools, uh students are better, you get it compared to, for example, my generation, but some are maybe even less educated in this because they don't really need to know about any building software. You know you know what is make cmake. That's that's what we have to start and then, even though don't know those you know flags for the compilers and they don't know about those as well.

A

Anyway, so that's interesting that might be um good for us to take into account with our new user training. Perhaps uh there needs to be a little bit of.

B

Background yeah background optional background training might be um yeah good, even helpful for pi. If those new members don't cause problems.

A

Yeah yeah, hey! Thank you cool. Thank you helpful idea, um so we're getting close to the end of our time. We have one last thing on the agenda. Oh and if others have got topic, requests or would like to you know, nominate to present something.

A

Post it post it in the slack webinars, I think, is a good place to discuss.

A

So, finally, for this week, um quick look at last month's numbers, so corey's scheduled availability was uh fairly fairly high. Actually all things considered sort of 96 um and we can see the yeah. The main unscheduled outage was the kind of the remaining time after the rather rather large c-scratch incident that we talked about last month.

A

And hpss and cfs uptime was all quite good.

A

um Corey utilization was uh 91.4 and 28 of jobs, um we're over the large job threshold. So that's um we have a uh obligation, I guess, or a target to support. You know 25 or more of our overall workload. We'd like to be you know, large-scale uh jobs that you know can't run on a on a system. That's smaller, pretty much! You know. Part of our mission is to support work that just can't be done anywhere else.

A

So our closed ticket slightly exceeded our new tickets in the last month, which has brought our backlog down to a little under 600. As of I think, that's a typo that should be the first of november.

A

And that's all I have for today. Thank you all for joining us, um we'll post these notes in the webinars channel and on the web page fairly soon and uh enjoy the rest of the week and we'll talk to you all more at the next one.