 Welcome back everybody to the sixth easy build user meeting. The first of our talks today is by Alan OK's from the Ulick Supercomputing Centre. Here we're talking about the Learn HPC project. Over to you Alan. Good morning everyone and yeah today I'm here to talk a little bit about the site project of mine or something that I'm interested in and which is kind of it's called I call it learn HPC right now at least. And it's basically about creating a scalable HPC training framework and general infrastructure. And if you don't know who I am my name is Alan Akash. I primarily work in one of the HPC European HPC centres of excellence. So that's called ECAM and I work for the Ulick Supercomputing Centre. So that's one of the biggest HPC centres in Europe and currently home to the number seven machine in the top 500. And they anchored the next talk with a little bit more about that. And so I just wanted to start a little bit with the with the with the European context. And so to talk a little bit about what's going on in Europe right now in terms of HPC and funding for HPC. So specifically I want to discuss the Euro HPC joint undertaking. So that's currently a group of 32 European countries signatories. And it's about developing a pan European super computing infrastructure. And they have this lofty goal that they want to have two machines in the top five the top 500 in its lifetime. The idea there is to support research and innovation activities to create the super computing infrastructure and also an ecosystem around it. And use that then to stimulate technology supply industry. And basically one of the big concerns is about generating European IP particularly for Silicon technology. And so the goal then they want to make these super computing resources available to a large number of public and private users and that includes SMEs. And for me one of the key words there is available. Right so so from what I can tell and from the way that the money is split the Euro HPC is primarily hardware initiative. And so they spent about 1.1 billion in between 2018 and 2020 and the budget for the next 10 to 12 years is around about 8 billion currently or at least that's a proposed budget. So, you know, as users of those kinds of resources what do you expect well it's probably not going to be that much different to how price operates right now. The resources will be open to everyone but of course it's a limited resource of the access will be competitive and based on your science and your capability to use them. So, you know, as a result, it's not unexpected then that of the 232 countries, those that have an HPC culture, what I mean by culture is that they have a history of HPC usage and development and it features a lot in among their scientists and among their formal education programs as well. Those countries that have that kind of infrastructure, they will of course have an implicit advantage right because in the competitive process, they're likely to be that little bit more competitive. And so talking specifically about this formal, the formal education stuff and again the European context here so the Commission, the European Commission commissioned a report back in 2018. That was looking at the academic offers of the formal education programs and also the demand for advanced profiles so the industry demand in the EU for AI, HPC and cybersecurity as well. And I just lifted a table from that report to show you here so like the numbers, I don't know whether that's good or bad but you know there's about a thousand courses related to HPC. So one third of those is in bachelors programs and two thirds in masters programs. So maybe that's good I don't know a thousand might be considered a lot. And, but you have to consider the context of the support right so this report was written in 2018. And for me at least something that jumped out straight away that was there was a kind of a big warning inside and, and it was related to Brexit. And so the context at the time was, you know, Brexit was happening and going to happen and it now has happened at this stage. And so in that report it was highlighted that the UK had about 62% of all the EU HPC programs, which sounds like an astonishing percentage. And, of course, there's an asterisk beside this because part of that was down to the methodology that was used in the report. So the report used an AI methodology. And it acknowledged that it was heavily biased towards English language courses so this may not be a completely accurate representation of the reality. Nevertheless, it's a big warning mark right about that there may be a for the EU for the EU what's now the EU without the UK. And this is probably a warning sign. And in some sense that's why I wanted to talk about like money. I've seen that in the national problems is where we're spending spending money and hardware is in some sense easy. And because you get something physical in the end that you can stamp aside and get your photo taken beside. But that hardware investment is is like it has a five year cycle. Okay, so it's not really a long term investment. If you invest in people that's usually it takes a long time to see the benefit of that, but the in the investment also has a much longer life time. But I think it's really hard right so bootstrapping the investment the hardware investment so that you see you come to see the impact in terms of the IP and and also the S the type of SMEs that you're generating. And that's a really hard thing. And it needs a lot of people, it needs people who are able to leverage the resources who are able to improve upon them and build upon them and contribute to these kinds of resources as well. Yeah, you could always import the expertise, but that's not really sustainable option certainly enough for for something the size of the European Union where there's more than 400 million people right so wouldn't be sustainable it's certainly not desirable. And also the something that you want to avoid in the US well which is internal brain drain right so having countries that are particularly strong in this case, we're pulling all the talent from the smaller countries, and who can't be as competitive right just because the size really. And so, what I feel is a space for is, is, or that there should be prioritized is a pipeline to be able to, you know, identify talent in this space, develop it and integrated into whether it's in the into the academic space or into the industrial space SME space as well. And, and you need that are as you need a very big pool of existing talent. And, and from my perspective I don't really think this exists. I don't think this is massive pool of existing talent exists always. We know about it right we wouldn't struggle so hard to hire people. And if this was the case. And so this is where HPC learn HPC comes in or the kind of idea behind it comes in the idea of to creating a big net to catch a lot of users at the at the beginner's level right so, and that's why it needs to be scalable right because we have to scale out to to a very very, very large community. And the argument that could be made in the previous slide is that you know you know there's always going to be good people right you're always going to have the cream who writes to the top. And, but there's, but you know you have to be careful right because you have to make sure that that you, you have a lot of people and to filter from right so that you put a lot of milk in the bucket in the first place right so that you, you put a lot of effort into making sure that you have a big group of people that that that contribute to the growth here. So, so me personally where do I come from so I've been like basically 10 years now working on EU projects that are primarily around developing and growing HPC user communities. I did that for five years in the geographic region right so in the eastern Mediterranean where I travel around to places like, like Turkey, Greece, Cyprus, Egypt, Lebanon, Jordan, and introduce people to HPC resources in those areas. And, and one of the most effective things we did then was was a road show we basically had a scripted, like two to three hour event where we go along, talk about the resources give people initial access as well so like let them log on to a cluster. Let them compile a code and run a code and it was all scripted and it worked really well for us we got a lot of penetration by doing that. I now work in an HPC center of excellence, which is like me to focus on material sciences. One of the events that we do is something called extended software development workshops, these are very different beasts, and we only do a couple of those a year, and they last between one and two weeks. Usually people come to develop a particular piece of software, but we take the opportunity to push the boundaries a little bit right so if they haven't been introduced to parallelism before we will take the opportunity to show them a little bit about that. Or if they're familiar with that we might show them more, you know, new technologies that they may not yet be aware of something like, for example, caucus to introduce some kind of independence from the underlying hardware particularly when you start talking about accelerators or something like that. I think in general that you have to be aware that if we're talking about EU level training and in this particular Euro HPC theory where there are 32 countries. And within that group, there are thousands, literally thousands of third level institutions and this is more than 400 million people probably. So, so basically everything that you have to do needs to be scalable, and to cast that kind of wide net. There are a lot of challenges right this, this for sure there's a technical challenge I mean the connectivity in different places is certainly not not it's not equal everywhere. And you've got like the hardware infrastructure situation that the software infrastructure situation and the expertise in the areas and then how it all that is all configured together so it's a real technical challenge. And but then you also have a pedagogical challenge right so so material right high quality material and instructor and people who are capable of giving that material well. And you need those people and you need it has to be that has to be skilled that infrastructure has to be scalable to because you need to go to so many countries and potentially in many different languages. And so what learn as we see really is this is just a matchup of other projects that you've already heard about some of them you've already heard about this week so yesterday we heard about magic castle, which is the one that generates this, this cluster in in in public clouds. And it supports, and the commercial providers like AWS is or where we get some of where we've got some sponsorship already, and but also open stack infrastructure so so the one that we use is the phoenix research infrastructure which is a another European project that provides some cloud resources for for for European researchers. And so we got some allocation of resources there. And on top of magic castle then we have the, the, the EC software stacks of the European environment for scientific software is installations that was also discussed earlier this week. And it's basically just, it's a software project to try and create a software stack that's common, and that are that can be used commonly across all different kinds of hardware, and all different kind of capabilities as well, and still is capable in terms of, well in the end it will be capable in terms of the accelerator usage and also in terms of the, the interconnects that are available. And as I mentioned here is something called HPC carpentry that I contribute to. So HPC carpentry is basically the software carpentry concept, which perhaps you're familiar with which is, which is a shared resource in terms of training materials so they create lessons and and less material which are detailed guides about how to give lessons on particular topics. The vanilla software carpentry program is on is on the shell and and Python basically. So the HPC carpentry one right now what we might primarily focus focusing on is an introduction to high performance keep computing but there are other initiatives as well on parallel parallelism. I've also done a lesson on running lamps at scale so that's another one that we have that I've been working on. But in general there are a couple of different groups are interested in the space and are creating lessons that potentially people can take away and use themselves. And the other thing that software carpentry does is an instructor training right so teaching people how to teach. And for me that was a really interesting process because, while I do, and give training courses pretty regularly. I don't do it every week I do it a couple of times a year so I don't have that opportunity to develop a lot of expertise, and just from experience and so learning a little bit about them pedagogical approaches and how how to be an effective teacher was was pretty eye opening for me that was really useful so I thoroughly recommend people to try and take that. And so basically we put all those together and that's this is the general idea behind what learn HPC is. And when I got the resources on the Phoenix resource infrastructure infrastructure one of the first things that somebody said was, Why would you use a toy cluster when you can give them the training course and the real thing. So my employer has some of the largest machines in Europe right and it's within my power to create training courses and allow people to give people access there. But that comes with an awful lot of downsides right so primarily one of them is that they're really strict security requirements right and lots of bureaucracy both for me as the instructor and also for the learners themselves. And this can be a little bit intimidating certain certainly when you're a beginner right so it's so people that asking you questions about things that you've never heard about before so like ssh keys or something like that. And, and I know that this is even more difficult on other sites as well. And then the other thing is, is if I'm running or working in the German HPC center where really consider running training courses for for a small university in Croatia that has no connection direct connection with the center. No, that's not a real possibility right I mean it's not it's not a realistic possibility I don't think. The thing about the cloud based clusters is that they're on a converter okay I mean I can bring them up can take them down I can throw them away. And, and also with the users themselves you can kind of make them aware that you know this is maybe not as secure as it should be like in the end I'm giving you potentially like just a username and a password and you can be on the system. And you may actually have the same password the same initial password as all the other users on the system. So they should know in advance right that nothing is there is to be considered secure right and it's all going to be thrown away in the end so it's not permanent either. And, but the other thing is is that while this is easy to bring them up take them down throw them away. They don't have to be toys. And so we've already successfully tested the magic castle infrastructure with the infinite band fabric that you can get on Azure, and also with the FA fabric on AWS. So we've seen like in interconnect results that are pretty comparable what we get our own sites. And you can also configure them in terms of the file systems they can reconfigure to run with scalable file systems right with a luster file systems well that's something that that physics and one from magic castle is also interested in having a look at. And, and of course the big bonuses that they're reproducible. So, in this case, what, and in this case reproducible really means scale up because we have a number of different events we can have clusters and that are actually events specific. So we can potentially run a lot of these, even at the same time, right for lots of different groups. So this is a really interesting possibility I think. And so what's the status. Well, learn HPC is really just an idea it doesn't have any funded manpower and what it does what we what I do have right now is a limited amount of funded compute resources so that and that's thank you to the Phoenix research infrastructure and also to AWS who sponsored me recently as well. And because they're interested in support on on the on the FA fabric. And the plan right now is just to check the relative stability and also easy itself is a developing project as well so there's not a ton of software in there but there is certainly for the use case that I have. There's enough in there right now to play with. And also I can add more if the guys are willing to take it on. And so I think these use cases can probably function already and I'm pretty sure that they work well with the stack. I'm still doing some testing on that. The plan courses for the next, let's say six months is to have multiple instances of this course from HBC, HPC carpentry, introduction type of forms computing, and then at least one instance of this running lamps on HPC systems. Because there's also GPU support in magic castle so so we just need that and easy first and and then other use cases as they arise if anyone has use case then get in touch and we can talk about it. And I was going to say there's a demo does not really demo in the in the chat I've I've given you details for for how to log on I set up a cluster for today. And so for the next hour or two I leave that cluster alive and you can log on take one user names that are there like trying to pick one at random and log on and have a look around and even submit jobs and try something out. It's an eight note cluster that's up there. So I'll take any questions now. Thank you to Alan for the talk if somebody has any questions then please raise your hand in zoom and will unmute you so you can ask it to Alan. I'm not seeing any questions now, but I think I've got one is if other people are looking at helping you, you know, where's the best place that they can contribute or So so so I really am relying on other projects right to put this together so so actually the bed the best way to do it to drive it forward is to use it, use it for a training course that would be great right so success stories. And because ultimately yeah it will be nice to actually have something like this funded right. And now like I say I have a limited amount of resources on on the Phoenix research infrastructure that give me more if I can justify the request. But to do that, I need to have success stories I need to have people who are actually using the infrastructure, and also to know like where the weak points I mean I found that, you know, things get updated magic castle gets broken and then fixed really I found is really good at that. And, but you know, to be aware of this, we need to we need people to be actually using infrastructure so so really just use it so get in touch with me and say you want to use it and we can work together. I see a second questions all on mute. Hi, I'm Sabri here. So regarding your comment about this funding for machines versus people. If you consider the HPC carpentries even it's mostly mostly volunteer right the most of the work that delivers this peer assisted learning is through volunteer work, although they get paid in other ways the real contribution come from people who really want to contribute not because of a salary. So how can you this competence center sort of strategy strategies could support these sort of initiatives. It's in simple words how can you play pay people who are actually contributing. Yeah, I think that's, that's, that's a hard one right so so Sabri mentioned competent centers right so competent centers is another thing out of you. Well, I think it's your HPC I'm not sure I certainly connected competent centers each each center as a national competent center competent center. And to be honest, I don't know too much about how they operate right now I know that that the way things were funded their people. Each country came up with its own plan so I don't think there's a, there's a pan European plan for competent centers I think there's a lot of national initiatives there about how they will operate. So I'm not sure it is unified thing, but there is a there is a pan European project, and that that links them all together and that's called Castile CST IEL. Maybe that's a good place to look. I really don't know that much more about them, but I do think that's a good place for for for initiatives that are about training a lot of people at a certain level right I'm thinking about the introductory type level or maybe you know a level or two higher but not at the level that price does the training courses. I think that price does good job at high level training courses, but there is space there for a more introductory stuff right and for supporting institutions and universities that want to do that, these type of courses themselves.