 Yes, so Welcome Yeah, so a little bit about me so I've been a computational researcher for 15 years in many different fields about five years ago I joined my current team and Now basically my focus is on usability and teaching of the computing So basically making it where people can actually use all the resources we provide So I'm part of auto scientific computing which is The name of our team within Alta University. We're a small team like nine people We don't really operate on a large scale But we are very close to the researchers Basically, we're distributed and we share coffee rooms with the people that use our services and it should note if you know Finland you probably know CSC we are not part of CSC and We don't even have as much interaction with them as we would like Basically, they're big and well We're not large enough to be to have a strong role there Okay, so what's the big picture? So I've phrased phrase this term crisis of computing. So basically The idea is that well people need almost anyone needs computing and researchers arrived They tend to arrive to our courses and our infrastructure unprepared for the work So an easy solution that some people can use to say well, you know, where the HPC cluster if you don't do real HPC This isn't what you need But that doesn't work for us. So we need to focus on everyone doing their work So we've got different things here. We have a training series, which you can read about more yourself We have a online course, which is more like a roadmap of all the different things You might need to know to build yourself up from what you might learn in academic courses to using the cluster So as I've said before Using a cluster is not that difficult when you know how to use the shell really well If you don't know how to use the show well, then well, it's almost impossible and So on and so on. So basically it's all about the basics here Lately we have a research software engineer service, which I think takes a complementary role here So not everyone needs to do everything themselves. So we can sort of export some of the problems to professionals who can work with the researchers so in Well a few years ago we started this kickstart course a few being maybe 2014 or so I'm not entirely sure and it's been continually refined over time So it's sort of a starting point of new users for a cluster. We recommend everyone to take it and These days it's not really focused on the high performance so much but on the computing part So most people if you need to use Large MPI programs or even write large MPI programs, but you can probably figure that out Yourself or use any many of the other good training courses that are available But the biggest problem is that people come and they start to use the cluster and you don't know Linux or shell or Whatever like that and you just can't get started So our typical contents these days is the first day is sort of general lecture-ish kind of things like a crash course about what does HPC and computing even mean About the Linux shell like what are the different workflows that are even available to do your computing Cluster or otherwise or guest talks from other people like CSC and then days two and three is how to actually use the cluster okay, um so When we got to a year ago and everyone everything went remote I immediately realized that this is a great thing because we're closer to everyone else in the world So it's easier to share things so for our summer kickstart course. We said well Why not open it to everyone else in Finland? so we did and There's different advantages and disadvantages here the Disadvantage is that the examples don't necessarily work out of the box everywhere else But we sort of worked on that by trying to standardize our clusters a little bit So it's not at the level of different courses for every different site But we tried to make one course that reasonably work between all of the finish sites excluding CSC For the reason stated above so the advantage is that even if we didn't have the effort to Generalize this to all the courses and make it seamless. It's still a resource that the other places just didn't have and realistically wouldn't have the staff to have in the near term and Most of the what I think important messages of the course are sort of independent of the cluster Like if you see how to run a array job somewhere It's not that hard to understand the concept and use it somewhere else If you know the Linux shell and other things to do the scripting So yeah, so last summer we did a joint with other Communities then in February of this year We took this a bit further and made it with the MOOC strategy, which I'll explain next and then in July oh Yeah, and then in July this year We continued that and really showed that it was seamless and scaled quite well So you can see our pipeline here. We had many people registered and quite high retention overall Okay, um So what's our current teaching strategy? So a lot like code refinery with the online courses Which is not a like small 10 or 20 person workshop like carbon trees, but we've managed to scale this to 100 people pretty well by distributing things into breakout rooms And then we have lectures, which is where we show demos and explain what to do and then breakout rooms for people work So in the Kickstarter course, we don't actually have this here So in the Kickstarter course, it's actually Like sort of the lectures and demos and typologs and then people working themselves with Mentoring and help when people ask for it And that's mainly because we can't get enough helpers like we can in code refinery So I think there's different levels of people we're talking to so some people are new and they're fine with just watching passively So you sort of need to like watch and see the course and see the whole thing And then maybe next time you can dive in a little bit more So expecting everyone to be hands-on is too much Some people are about ready and want to do all the examples and the courses about on target for them Some are quite advanced and want to move forward quickly Another important thing is co-teaching so we basically say you should never teach alone So in the small courses the idea is that you have the teacher and the learners and the teacher should have a discussion with the learners but that doesn't really work in the giant courses so in code refinery we have a hundred people and You can't like basically the audience didn't talk with you So instead we have two teachers and the teachers are talking with each other And you can read here if you would like to know how it works in practice And this basically makes it seem interactive and work very well Then we have hack and D for questions, which I forgot to introduce but I guess I can introduce now So here which I will put in the zoom chat is The hack and D and you can ask a question such as like this and then you will get answers like this and And this is basically better than voice because multiple people can talk at the same time And you don't have to try to speak over each other in order to like get a time to ask the question and it's better than the zoom chat because Well, it's sort of asynchronous and you can go back and answer in different orders So this basically lets any course scale and become super interactive Okay, so the tech setup this is where it gets interesting So when we're teaching to 200 people We don't do that in one zoom meeting first off. There's no need to and it doesn't really work so The basic idea is this there's one instructor zoom meeting here Which you have the instructors and some certain helpers and the production crew and so on and learners are not there Instead it's captured with open broadcasting software Which is what I'm doing now and then it's live streamed to the whole world via twitch so basically anyone who once can sign up for it and Watch and see and there's no risk of it affecting anyone else I mean, isn't it sort of interesting that as soon as we go to remote work We have the possibility for anyone in the world to attend our events and we immediately learn Okay, we should never ever share a zoom link because something bad might happen Why not have both this provides that? so and then there's a learner zoom and the learners independently watch the twitch stream or it can be shared within the zoom meeting and then there the learners work and Like do the examples and talk with the helpers and mentors and then give us feedback via this HackMD and then we see that in respond and Comment back on the stream and so on so the general flow goes We'll talk a little bit will give an example and say okay now it's your turn and then we say okay you'll have 15 minutes and then we announce it we mute the stream and then the stream goes off and Can work alone in this? learner room and then We since the stream is muted we have our instructor space to talk and plan out the next strategy without interfering with all of the other Learners, which is really great. It hides the back channel talk some So as you notice here, I'm sharing zoom with a portrait screen share. I think this is a very important thing so It basically means that we have half the screen for us and our examples and half the screen for people to work along and Do their own thing? Yeah, so even in this meeting You're probably gonna be you're probably doing something else at the same time. You're listening to me So why not make it easier for you to do that? So it's easy to give the attention to me instead of minimizing this Landscape screen where everything's too small and you have to make it full screen to fit anything This is really great So a typical learner layout would look like this here Where half the screen is the share and half the screen is available and Well, we mix in the Instructors talking via the OBS open broadcaster software. So what are some advantages of the setup? so Well, you can see one of the feedbacks we got from some users here and basically Well, I like it a lot and it seems that some other people like it a lot too I see little reason to try to go back to another kind of strategy and I even think it's so Good that I don't really see how we can get something. That's as useful within the When we go back to in-person courses Let's see and Yes, I will publish the questions and answers later if there's anything useful there Okay, so what are the advantages so HackMD is a much better tool for answering questions and Oh, yes, so Yeah, so the HackMD since there's no personal data there can be shared and we put it on the course website Or make it available also since the learners are not in the Not in the same zoom meeting there's zero chance of privacy risk here, so we can immediately release the videos We can allow anyone to watch without giving out this zoom link publicly Let's see The videos are available for an instant replay So twitch records it for 14 days and then it gets on YouTube soon Which I mean when you have a course like this everything you do in the course I mean that's not really enough like do we really expect someone to learn what we're teaching in the three hours we have in one day, so No, I don't think that's really possible. So we need to mix these different strategies There's a live thing where we talk. There's the written material. There's the recordings later And this also makes it really great easy for someone to catch up if they get behind The instructor back channel doesn't interfere with the learners So in the original code refinery courses when we're going in breaks or exercises and then instructors are discussing Oh, what do we do next now? It's great We can sort of just mute it and then we plan ourselves and the instructor workload is much less People say that the team teaching is much more engaging than trying to get these overloaded learners to Get Get engaged in it and I think the course is just overall a lot more fun to give So what are the disadvantages? So well when you try to scale too much for something like HBC, there's problems with incompatible sites So we got a lot of feedback about this but when discussing it then other people come up and say yeah like and maybe it doesn't work but This is still much better than anything we had so just keep going at it So the tech takes some getting used to both for the other instructors and for the learners Yeah, but I mean everyone got used to it pretty well after you saw a few examples so I think that this is not really a big concern and When we went to remote work, I took some time to upgrade my home desktop computer and made it where it was able to Like prepared it for these kinds of things and as you can see here, it's slowly grown into this massive setup with the Let's see So with the this is the code refinery setup and it was a bit more complicated with the current setup You need one less screen to do it. Well, but I think you could even do it with two screens or something like that And really this is to the point where when we go back to the office. I don't think I could run this from work I have to have my home setup in order to do this. So what are some future prospects here? We'll probably keep going with this strategy for our large courses at least the really biggest ones And there's really no reason not to invite the whole world to attend Other people want to take part they can simply provide the Twitch link and they can say okay, you know, you can watch this course it may be useful It may not be you'll probably learn something you'll have to go and use our docs to fill in missing gaps But at least you get the big picture You can make your own zoom meeting with for your own breakout rooms So basically as one of my colleagues said it's like a TV production So we have the studio we produce it for the world and then there's all the different people that are earned at different theaters or classrooms or audiences and then they watch it separately and then You know during the commercial breaks then say okay, now let's talk among ourselves, okay now Let's go back and so on and so on also, I think rather than making the courses different or Customizing the courses different things why not try to standardize clusters more of course this is a almost impossible task because of the independence of Sis admins of different sites, but you know, maybe we can make a generic course which has the basics and Say okay, we expect that Your cluster can work with this course Let's see other Things in the future so Yeah, it's the course generic or localized um Do we Continue sort of like we had where we had some generic days and then some Days where each site has their own program running on Can we combine more internationally is code refinery platform for this? These are all good possibilities here So let's look at these questions here while I had some other last-minute things they wrote down So yeah, I think that the usability is as important as the training if something is hard to teach then maybe we need to make it more usable Let's see So this is a part of the big training courses and also I think that the courses are not the end all Strategy here. So I really like I said above I think I really don't think that a single course is enough to have everyone learn things So instead of forcing a whole course into a few days and expecting this is enough I think we have to have a course which is reasonably good enough and scaled enough Like sort of what we're doing here and then good local support. So the person is really going to learn HPC by By asking their colleagues or their local support for mentoring in these like Small little things that they're working on as you're working it on Okay, let's see What's the backup plan so well so our backup plan so far has been if the Streaming setup breaks then the instructors joined the published zoo meeting for the learners with the breakout rooms and then teach there Let's see Let's see Yeah, so I think that we need to Make this where other people can use the setup pretty easily So we're not quite there yet in part because I'm still refining it But I think we're slowly getting there and when someone has time they'll be able to do it without too much effort Yeah, Sima had a point here about Like it really does sort of centralize most of the tech setup to one person And that actually means that we can take a lot more instructors in so specialization is okay Yeah, how does course evaluation work so by having hack MD which is continually having questions like this then The instructors can get a continual pulse of how people are asking questions and how it's going We can have feedback We can say okay, how would you like how does this feedback work? Like how's this going and we ask people to give the feedback in the hack MD and everyone can we have polls via hack MD and so on and My time is running a bit low here But Yeah, I think that's the general point. So I guess in summary I sort of see that The goal with the teaching should not be to take the training courses We would do in person and make them online But sort of have a new strategy like we have to have a whole new concept of running the courses online and then we can reach far more people and fill in a gap and then we redesign how we learn and Mentor in person and small So what's the instructor participant ratio? so in the Kickstarter course we had 200 learners or so and then maybe Maybe five to ten helpers on a typical day during the course within the Learner room So these were answering questions from learners and we would have a few from our site a few from the other sites within Finland And then for instructors there was basically me and then one or two other instructors for most of the time So the scaling is just amazing here Yeah, but well the downside is it's not a One-on-one thing like the Carpentries does So okay with that being said I will stop my Screen share here and go back to zoom so Yeah for anyone that's watching the recording later. Thanks for listening