 So, welcome to our virtual KubeCon 2021 presentation. Thank you for coming. Well, thank you for watching online, as I should say. We're going to talk to you today about the CNCF Research User Group, who we are, what the group goals are, who comprises the group, some highlights from recent sessions we've had, and also a survey of our constituents. Next slide, please. So, yeah, the group is run by myself and Ricardo. I'm Jamie Poole. I'm an engineering manager at CG Research. I manage a team called Compute Platform Engineering. We're responsible for all of our Kubernetes Estate button cloud-native technologies at GR. Ricardo? Yeah, so my name is Ricardo Rocha. I work at CERN as a computing engineer in the CERN Cloud team. I focus mostly on containerized deployments, networking, and more recently, accelerators and machine learning as well. So, we've been using Kubernetes for quite a while now. So, and we engaged in this group. When the idea came, I think it was KubeCon Barcelona in 2019, I guess, early 2019. So, yeah, it's been a pleasure to work with Jamie as well and everyone else. Cool. So, we'll talk quickly about the goals of the group here. So, as Ricardo says, I think we put the group together at the beginning of 2019. I feel it could have even been earlier than that, but it was definitely the KubeCon in Barcelona. There was a bunch of people who came from different research institutes who wanted to use cloud-native tech and wanted to solve the same kinds of problems. So, we sort of came together, I think actually over lunch. And then from there, it sort of became a proper CNCF group. And yeah, basically we want to just bring together people from different research institutions who want to solve the same common problems using cloud-native tech and really build a sort of research focus community who know how to deal with the same sorts of things that we will have to deal with, maybe slightly different from your traditional cloud-native technologists, I suppose. We ran a poll. In fact, no, this wasn't the poll, was it? This is actually from our weekly, bi-weekly, sorry, meetings. We keep track record of who attends and who speaks. And this was the results based on, I think, a notebook that Ricardo put together, which he's going to show you the gory details of in a minute. But you can see we've got quite a wide spread of institutions from around the world. So, I'm from G Research, so I often get a lot of us to come along. That's why our name is so big and central in there. So, a big constituent part as well, always present. But yeah, quite a wide range here of different people from different universities, private, public institutions from all around the world actually, literally from the, all the way from the UK to Australia and back again. So yeah, really great group of people to get together. Yeah. And it's even good to see, like, not only individual institutions, but in some cases we see, well, I work on the physics side. So we see institutions like NFN, which are actually very large groups of different places in Italy or EGI Foundation, which is driving the grid infrastructure in Europe. So yeah, it's been quite a nice set of diverse groups. And as Jamie said, I guess, because this is a research user group and well, we do a lot of data analysis or at least the users of the services we provide do a lot of data analysis. So I thought it would be fun just to collect all the attendees from the agenda page and do a quick analysis on the list. So two very silly ones in addition to the world cloud that we just saw. We got 65 unique attendees over the last 10 meetings. So it's not like the whole lifetime of the group, but just the last 10 meetings, which is quite good. We had some sessions, I think, with over 30 people. So it's been reactive. And then also the different institutions, we have 34 different institutions participating. So it's been very, very active. And if you're interested, I put here the link to the notebook as well. Yeah, so highlights here from the recent sessions we've had. The structure of the group at the moment is we meet on the first Wednesday and third Wednesday of every month currently 4 p.m. UTC. You want to go try with times that's changing. The structure typically is we just do intros. Anyone who hasn't come to the group before does a little intros of themselves and says hello to the group and we sort of greet them. Then ahead of time, we've decided on a particular topic we want to talk about that session. We'll have a speaker who typically presents the chosen topic for 15, 20 minutes, normally some kind of presentation, normally a little bit interactive if possible. Got a screenshot here of Jeremy from Oak Ridge talking about their really cool work they've done recently on POSIX integrations. Yeah, then Q&A for 10 or 15 minutes and then a bit of wrap up, talk about the topics we want to do for next time and then move on. We record all the sessions. They're all available on the GitHub page. I think actually shout out to Cheryl Hung as well who organizes it for us. She's in fact going to livestream the next one on YouTube and we're going to try that out for a bit, see how that goes. That should be really interesting. In terms of the recent topics, they've been quite diverse. Obviously all research focused. Things that our members are interested in. Mauro from G Research did a really interesting talk on OODC a couple of weeks back. I think Ricardo, you put together the notebooks one that was very well attended, lots of interest there. Do you want to cover that? I think three different institutions showing how they are providing notebooks on Kubernetes. Things like Tube to Hub or Binder or in some cases using tools and frameworks like Dask for Python. That one was a pretty good one and actually triggered some follow-ups between the people that showed interest in those and we started collaborating on creating a recipe. We'll talk a bit more about that one as well. There's a couple of other ones. I would also do the shout out to Cheryl. She's been pushing for this group quite a bit and helping us with all the recordings and all the setup. Also Bob that actually founded the group at the start. In terms of topics, I would also highlight two that are very close to our needs as well. The first one would be a rootless Kubernetes and this is very important for all those institutions running HPC clusters where you want to make use of some of the cloud native tooling but you don't actually have root access to the nodes so you need to deploy things in a different way. So it's quite important that all the components not only Kubernetes but Continuity and all the rest of the stack supports rootless mode. And then the other one is image distribution because we typically do very large submissions of workloads and being able to start these jobs quite fast is very important and also maybe it's the case for everyone. Researchers are not necessarily very good at layering their images so we often end up with very large images. So we had a really nice presentation from Akihiro Suda for rootless and also from Ktalk from NTT as well talking about the Continuity Snapshot and the StarGZ support. So these are really, I recommend that if you're interested in these topics it's a nice way to start to just watch the videos from the sessions. They're all available in the agenda page. Do you want to pick up, Jamie? Yeah, so we were thinking about what to do for this talk and how best to sort of convey what the group's all about and what our interests and pain points and domains are. So sort of use the technique which you used in the past which is actually really effective which is just to put a survey together with some pointed questions about what the institutions do and what their main challenges are. Send that out and thank you to everyone who did reply who's already in the user group. And in the next few slides we're just going to go through some of the results which are quite interesting actually I think and some of the answers a little bit counterintuitive to what even we might have thought we would get back from the group. So yeah, some quite interesting data here. Yeah, and we'll keep the survey open so it's still time for everyone to jump in and then we can do an update later as well. But yeah, we can start just with an overview of the institutions that already replied. So we can see quite a span of different areas already in the names. There's a couple of cases where there's more than one reply. I know for example for CERN this happens we actually have different groups at CERN that focus on different areas. And yeah, everyone got the notification so we ended up having more than one person reply and I guess for all creatures, a similar situation. Can I go, Jamie? Yeah, so no surprises here I suppose the bulk of it is sort of non-private sector anyway. I think private sector was just represented by me in the feedback that we've got a lot of academia, a lot of government, a lot of nonprofits when you add it all together. So yeah, it's sort of I guess not particularly surprising in the sense that that's the way a lot of bulk of the research happens in the world. I expect there's a lot more private sector out there that aren't represented in our group. So these are definitely some people I'd be keen to sort of reach out to with this presentation and try and encourage to join the group as well to give their views as well. Yeah, I think there's quite a... The research use case is like... At least for someone who often associates them with research laboratories and academia but actually there's a lot happening in private sector with similar requirements. Yeah, this one was just to have an idea of the size of the institutions so we have I think probably one answer over 10,000 people. Most cases are between 100 and 1,000 so it's still pretty large. Yeah, and quite a few between 1,000 and 10,000 as well. So yeah, we're talking pretty large scale requirements here. Here we've got the different areas of research which people primarily support. So quite strong showing from physics, bioinformatics and generic research platforms. I guess there's potentially a little bit of murkiness as to what's going on in there. But yeah, the bulk of it I would say is sort of science and scientific research I would say from this feedback. Yeah, internet tool also for more education. It's quite interesting. And astronomy, there's one community that hasn't been very present which is from the specifically the SKA but actually from astronomy we have quite active members from ALMA from the observatory in Chile. I think we should get them to do a talk at some point even to show off their nice telescopes. That would be very cool actually to see what those are up to. Or just to organize a meeting in the middle of the desert in Atacama. Yeah, I completely agree. Okay, so this is really good to see. Very, very clear that the vast majority of the group are already using Kubernetes use in production. So this is a group of people who do really understand the platform and are really using it for value add and getting something out of it. This is not just a sort of collection of people who are hobbyists or trying it out for the first time. This is really sort of proper production use at scale. Certainly I know a big scale users and so we actually saw very large amounts of compute growing on Kubernetes in production. However, that being said, there are some groups here who are sort of just starting out and just learning. So it's also a good way for people to get in and learn more about the platform. See what it can do for you. Yeah, it's even more surprising because, yeah, we'll see also in the next answers, but there is a kind of a learning curve to getting to Kubernetes, both on the infrastructure and the usage. So it's quite nice to see that so many people are really using it in production for research as well. And then, yeah, the other question we asked was which kind of infrastructure is being used. So I'm actually quite surprised that there is such a high percentage of on-premise deployments, but I guess it's quite specific for this community because we have large data and large-scale deployment requirements. So I guess people already have this infrastructure on-premises and they just converted to use Kubernetes and Cloud Native. But I guess for industry, this plot might look quite a bit different, more tendency to public cloud only in some cases. Yeah, I think you're right. I think certainly for the sorts of institutions that we tend to deal with, these have been sort of large-scale HPC research shots for a long time, long before Cloud Native tech existed, I suppose, so you probably already own large data centers and amounts of hardware. So, yeah, certainly for us as well, it's a case of how we use that in a Cloud Native way without necessarily jumping straight into public cloud. Yeah, and the follow-up to this would maybe, to this survey would be to understand better the requirements for hybrid deployments as well because I know in our case, we are looking to extend the on-premises resources and try to fish for things that we don't have too many, like GPUs and accelerators. So it would be nice to understand not only why, but also how people are doing this. Yeah, absolutely. Yeah, we're just looking at the why people are, picking up Kubernetes and using these Cloud Native tools. I think it's kind of interesting. A lot of this, saying the latest last as well, I think is really, we're just sort of highlighting the benefits of Kubernetes, but really showing that you can actually make use of those benefits in a research setting. So you can imagine in research institutions, things like reproducibility and fast iteration are super important because you need to be able to rely on your data. So that's why certainly for us, I think, things like tools like Kubernetes really, really pay off. Yeah, definitely. I think it's pretty clear, like eight out of 12 talk about reproducibility. So that's a big one. It's a matchless, but multi-cluster, which is kind of interesting. So that's something that's worth following up as well. And then this one surprised me at least, that so many, so many deployments or half of the deployments actually more if we count the next, actually provide direct access to the clusters, to the end users. But it's nice to see. Yeah, it is a big move forward to do this. But yeah, traditionally, at least in our deployments, we already have systems on top that will do the higher level, expose the services to the users in a higher level way, not really necessarily the Kubernetes API, but we do have some services like this as well. I don't know if that's your case as well. Yeah, we definitely have a mixture. Our researchers don't necessarily directly access Kubernetes all the time. They have a bunch of tools to do so, but then there is a sort of mixture because then they do to sort of look at the gaps of it and see what's going on. So I think it's good to be able to trust people to do that and to be able to set it up so that they can easily with a little bit of education, I suppose, of how the platform works. Because I think, again, we said it before, we'll come to it, but it is quite a complex platform and allowing people direct access also implies that they need to understand a sort of baseline level of what's going on. So maybe some interesting topics. We already covered OIDC, but going deeper into how people are using our back and things like the Open Policy Agent might be things to follow up as well. So here we just, I guess, threw out a bit of a list of popular cloud-native projects which are relevant in the community today to see what the sort of spread was of people across our group using these things, and no surprise Kubernetes out front, almost 100%. A lot of Prometheus, pretty standard, I suppose, of collecting metrics and looking at your data. I noted that service mesh was particularly low here. Chatham Trikala just before this, I think probably actually in a research setting that's not maybe that surprising. It's less about services and more about batch and HPC and that kind of thing. But yeah, a good spread of these different tools. I'm also very interested. One from the 12 responses is not using Kubernetes, which is kind of curious here. The other two that are interesting as well is Ardo. I guess it's gaining quite a lot of popularity also for managing workflows. So that's an interesting one. I know tools like Kubeflow also rely on it and there's others. Yeah, I see these come up a few times as a topic to talk about as well, to see how people are managing their states because there's lots of different models. Even me and you do things differently. You tend to provision clusters for people to then use and they own them. Whereas we do the more managed approach, but both approaches you need quite good reliable CICDs. So yeah, it's very interesting stuff. Cool. Then what types of workloads? I guess this is, it's pretty amazing that 10 out of 12 are doing Chukta Hub and interactive analysis type. I guess the popularity of these sessions in this specific session in the group kind of also showed this. And second is batch. So I guess these are clearly the two top things that the members of the group are focusing on. Yeah. It's kind of surprising the amount of people doing databases and Kubernetes, but I guess, yeah, why not. It will come more and more, I'm sure. I don't think, I mean, we've started this, and we're not doing databases in production yet, but I think over time and sort of consistent storage tech gets more and more reliable in this space. People run more stable services. I don't see one. The one curious, one thing that we've been trying to do is offload our central services, especially for the storage side and provide in cluster storage in addition to the persistent centralized storage in a way to have, for example, if you're running large workflows to kind of have offload the temporary storage from the central services. This is something we are looking at as well. Yeah, and then we get to the two main, two last answers, and this is kind of the main benefits. Clearly here, there's nothing really that jumps out. It's all pretty similar to the general benefits from Kubernetes. It's a nice list here. I think it's all covered. And then the last one we have is the single main pin point. I don't know if you want to highlight something, Julie. Yeah, I mean, I think the two ones that jumped out for me was the different mental model, steep learning curve along with overall complexity. I think we all know it is quite a complex system. And once you get over that sort of initial learning clip, you can really start to, you know, maximize the benefit from it. But I can see that for people coming on board, it's definitely quite complex and there's a lot going on. There's other points here, which are designed to that around pre-packaged software and integration with other systems. But yeah, it's clear that it's not completely smooth sailing for people, but I guess that's why we come together in these sorts of groups to work out how to make it smooth sailing and becoming these problems. Yeah, the pre-packaged software is quite interesting as well because for most generic tools, it's probably already there. It might just be that we are lacking something on this specific area for research and scientific. And then the other interesting one is actually reproducibility because it was one of the, like, top reasons to rely on Kubernetes and Cloud Native, specified as one of the main points for using Kubernetes. So it's clearly work to be done here still. Yeah, so then the last bit is regarding where we can help the most. So yeah, I think it was, the goals of the group are also shown here. So it's increasing the visibility of research use cases, but then there's some specific things. We saw in the previous answer also this need to help in package catalog and software. And then the other one I would highlight also is this idea to maintain a set of recipes and best practices for use cases we discussed. And so I think this is something that is already happening or starting to happen, but it's true that we haven't had any result yet out of the group for these kind of recipes. Yeah, I think there'll be some real benefit there. That's definitely something that will address some of those pain points. Just having something that researchers can kind of pick up and use without having to do too much thinking. Yeah, when we discussed the Jupyter Hub, actually, there was kind of a lot of commonality, but a few differences also on how people are deploying their workloads. One was also, when you deploy your interactive analysis and then want to offload to an external system, what is that? And there was discussion about how to integrate with things like HDConder. There are other discussions on how to integrate with Kubernetes itself. There were other solutions. So maybe this is something where we should focus on it. Yeah. I think that this is the very last one. Yeah. Just some comments here that people have for suggestions, but generally that's how people are enjoying the group. It's good to see that people appreciate that the group exists. So we're very happy to keep it going and just want to encourage people's collaboration. I think the next slide, just talk about our plans to build on the momentum we have now, collaborate on these recipes and just really engage with other groups as well, try and get people on board and solve some of these problems. So yeah, I think the final slide, if you move on to how to get involved, I think it's probably enough of me and Ricardo talking, but yeah, never look. This is our GitHub page. It's got the details of the Zoom meetup which we do currently first and third Wednesday every month. If you're on CNCF Slack, it's a public group UG research. And yeah, thank you everyone for all your participation over the last couple of years and let's keep it going. Thank you and see you soon. See you soon. And yeah, I think we've got some time for questions now. Thanks.