 Okay, so our next speaker is Yusie Enkovara. So I have already made a couple of links to the CFC during my presentation on the very first one. So Yusie will tell about the real computers. So he is originally from actually at that time that used to be TKK. So to those of you who is at Alto, I don't know the history. So here is the kind of history lesson for you. So Alto was organized in 2010. But before that, that was TKK and along Yusie, as myself, we've got also the PhD from the Laboratory of Physics under the supervising of the great professors like Marty Puska and Darystony Eminem at that time. So if this name tells you something then. And nowadays, Yusie is working for CAC, Specialist in Scientific Computing or Specialist in HPC Computing in general. He was also helping us at some point. And nowadays, I mean, he's a really experienced guy. If you have some quote anything with respect to the agency, then you'd better ask definitely him and use your opportunity. So Yusie, the floor is yours. Thank you, Ivan. So let me share my screen so you can hopefully see my slides now. As Ivan told, I come originally from Alto, as I said, here is the University of Technology background in. I did my PhD in Atomic Scale, Quantum Mechanical Simulations, got bit involved in programming at that point. And I have been working at CSC almost 16 years now and I work with very many different kinds of supercomputers. So within this presentation, I tell you a bit about what kind of services CSC provides for researchers. Of course, we have the supercomputers, so I talk a bit about them. But there are also some other services that might be useful for you in addition to just the computers. For the questions, etc., I think we use the same HecMD as previously. And actually to start with, I put a couple of questions there. So you hopefully see the short poll there. And there are quite a bit of slides, which you can, of course, you can access all of them. I might not be able to go through all of them, so especially these. So just spend more time in technical details or data management services that would be interesting to know. And of course, generally, it would be nice to know whether you have used CSC services or maybe some cluster in the FGCI. And let's just wait a couple of seconds still to get this actually quite cool idea using the HecMDS as a sort of bar graph like this one. Okay, I think that's probably enough. I think I'm not going to do two details how to actually submit the batch job in this presentation. I think that's something that will be discussed more in the other sessions. I'll just maybe mention already this point that the way you use Triton or many of other clusters in FGCI is actually very similar to the CSC, so in that way the transition is quite smooth. For those of quite a few of you will say that you haven't used CSC services actually. I'm 99% sure that most of you have actually used CSC services without knowing that maybe not to computing services, so I'll discuss also a bit of the other things. But as I said, here is the brief outline what is CSC, then I think there was interest also about the data management services, so I tried to spend some time on them and then of course some support services and finally briefly that how you actually can get started using maybe more with the computing services for this last point. Those of you who are not familiar at CSC at all, we are in principle a private company owned by state, the Finnish Ministry of Education with the 70% of CER and then all the Finnish Harjo education institutes, meaning universities and the polytechnics with the 30% CER. And even though we are sort of a company, we are a special company as this is, we are not making any profit, so we are a non-profit company. And tasks of CSC provide various IT related services for science and education. Very super computers are just one part of that. Most of us in normal times, of course now everybody is more or less remotely, we are in Keilandiemi, so quite close to Ota-Mieni campus. And then we have the data center where the super computers are in Kajani. And nowadays we are more or less 500, I think with the summer studies starting now, I guess we go above 500 people. Probably one of the most important things for you to know is that most of the services CSC provides for individual researchers, they are actually free of charge. And really meaning that you do not need to pay any money for that. Of course somebody needs to pay for this money, basically it's a Ministry of Education and Culture. They have made the agreement that most services are something that the researchers or even the university do not need to pay anything for them. There are some sort of limitations what can be done. So if there is some commercial usage of services, if you are doing a R&D together with some companies where the research is necessarily not public, then this might not be a free of charge. You can find a bit more in our web pages about these more special cases. But I guess it's for most of this audience, if you want to use CSC services, if you want to use the supercomputers, it doesn't cost you any money directly. As I say that most of you actually, I think have been using CSC services. So if you ever use the Internet at the University of Kampus, CSC is operating the Internet network between the universities. If you ever use the Eurom wireless in anywhere, that's in principle also service provided by CSC. So CSC is sort of contributing to service and also many student administration systems and so on. CSC is actually doing work also there. So I think if you go to Opintopolku, et cetera, many of these systems, CSC is actually somehow underneath there. If you use Haka authentication, that's in principle also service where you are implicitly using CSC. But for this presentation, I'm mainly focusing now on this computing and software, a bit about data management. And let's see if I can... Data management is a bit about support and for training, et cetera. Customers, main customer segment is researchers in universities and also in state research institutes. Then something more for education and then some government organizations, for example, the Audiovisual Archive, Finnish National Library and so on. They are also using some of our services. Okay. And yeah, then maybe also mentioned that you will sort of... We will be providing... Or you can get support in many different phases of the research. So starting from the planning of the research project to actually running the simulations in our supercomputers, maybe analyzing the data. Then finally, sharing and publishing some of the data sets. Okay. Let's see, somehow I managed to get this annotation there. Let's see if I can just create a drawing. Looks a bit nicer now. Maybe also some sort of things that are done now are going to give a bit more perspective. So there are nowadays, there are many different types of scientific problems that can benefit from supercomputers and the kind of infrastructure that CSC provides. There are of course the very large scale simulations where you need hundreds or thousands of CPU cores to perform the simulations, which are really the big ones that you don't have any similar resources in anywhere else in Finland, at least. Then there might be some smaller scale simulations where the number of CPU cores for individual simulations might not be that large, but you want to run lots of them. Many of the data analysis tasks, machine learning and so on, they sometimes can benefit from CSC services. And maybe a bit more, more special case and I don't know how relevant for these audiences, but for science dealing with sensitive data, so for example, some biomedical research with the patient data, et cetera, through a special services for handling this sensitive data. And as the sort of new type of services for also for experiments where blood amount of data is generated, so you might have some satellite or big physics experimental apparatus that produces huge amount of data and these can be in some cases semi-automatically transferred to CSC. Storage and then analyzed with our supercomputers. A couple of examples of research that has been done and published in supercomputers, different machine learning based methods, they are in many areas, they are very important. So there was some applications done by Pekkarusvuri. Nobody sees actually I think in the University of Turku, but they use CSC computers for making diagnosis for cancer. For sort of coupling different length scales and different types of simulations. Many climate research is a good example. So there was from Professor Mariko Kulmalen University of Helsinki, they have used lots of CSC services for studying atmospheric feedback mechanisms. A very timely example I think that was studied in about a year ago is studying how aerosol particles spread in air. You might even see some simulations in YouTube and in TV, where supercomputers have been used to study. So how for example coronavirus will spread in the air via people inhaling etc. Molecular dynamic simulations of a biological cell and so on, they are some very sort of common type of simulation. And that's maybe illustration that really one big benefit of supercomputers is that you somehow you can think that it's a complex scientific apparatus, which seems to resemble in many ways the laptop you might have. But you can still think sometimes that it's a complex scientific apparatus, but it's something that you can with the same device. You can study either the smallest things in inverse subatomic particles, but then also things in cosmological scale. So here is sort of type of simulation related that with the gravitational waves and stuff from early universe. I studied with the Putsupe computers and this is what this particular example that was using over 10,000 GPU cores for performing these simulations. Here we can see the overview of the CS infrastructure. So as Ivan already mentioned in his presentation, we have two main supercomputers at the moment, Pufti and Mahti. I discuss a bit more details about them in a moment. Both of them have both CPU partition, so it is normal Pufti and Mahti. And then we have also part which has GPUs. So that's actually Mahti that's relatively new. That was in production only a bit more than one month ago. These so called Mahti AI and Pufti AI parts. Both of these supercomputers, they have what they own disk systems, fast parallel storage for very efficient input and output of massive amounts of data during the simulations. And as a common storage system between them, there is the ALLA's object storage. And nice thing about ALLA's is that actually the data that is there, that is also, you can make that directly accessible via internet. So in principle, you can access the same data from your laptop or from your from CSC supercomputers. And you can even sort of get the URL for the dataset you have there for easy sharing between collaborations. We have also various cloud services. So we have AIPOTA, which is cloud service for sensitive data. That actually if you want to use that, that requires a bit special arrangement because specific network connection needs to be made there. Then we have a C-BOTA and then we have also container cloud. And if the CSC resources are, let's say the current CSC resources are not enough, you CSC can help you also for getting access to some big supercomputers in Europe. We have the base research infrastructure. And soon we are getting one of the biggest supercomputers in the world, Lumi, which I tell you also a bit more details a little bit later on. Puhti, that's our sort of general workhorse. So it's based on Intel CPUs. And it has a range of memory sizes for different types of simulations. There is also a bit different local storage options. So some of the notes, they have a faster disk and all the notes they're connected with the quite fast interconnect. And it really meant for many types of simulations from interactive single core data processing to some simulations which can can utilize at least hundreds of CPU cores. The Puhti AI part that consists of 18 nodes with NVIDIA P100 GPUs. And it's especially for various deep learning frameworks and so on. But also HPC codes that can utilize GPUs can use that. And it has a quite large selection of scientific software installed there. Puhti is really designed for massively parallel simulations. So normally the smallest unit that you will be using with Puhti is a single node. And its node has in total 128 AMD CPU cores. So there are two 64 core CPUs within a node. But it's more homogeneous in a sense that all the nodes they have the same amount of memory, same type of local disk. And I say there's an addition. There is now also the part with GPUs where we have 24 GPU nodes. Each of them has four NVIDIA A100 GPUs. So they are a bit more performant than the ones in Puhti. Okay, sick. What do you have had any questions? There was one interesting question. What does CSC stand for? Yeah, nowadays CSC doesn't stand for anything sort of like GNU. It doesn't really, I don't know if it stands. CSC, it used to be Center for Scientific Computing. So I think that that word comes from nowadays. I think the official name of the company is just CSC IT Center for Science Limited. Yeah, that's how it comes. Okay, a bit more to the technical details then. So this is pretty much what I already mentioned. So if you look about the technical specifications that we have in Puhti. So the CPU model, that's Intellexion Gold. You can find the actual model number there. And there are two CPUs per node. Both CPUs have 20 cores and they run with 2.1 gigahertz clock frequency. As I said, there are different memory configurations there. So as you might know, sometimes memory might be even more expensive than the actual CPUs. So when trying to make compromise for a system that you want to get lots of CPU performance, but also support some cases where you really need lots of memory, you need to sort of make, you cannot have the huge amount of memory for all the nodes. So most nodes, they have in total of a bit less than 200 gigabytes of memory. But then there are a few with the medium amount. And then six nodes, which really 1.5 gigabytes of memory. So if there is a simulation case that some quantum chemistry applications using the way function based methods might sometimes be an example which really need lots of memory. They are also available there. Most of the nodes, they don't have any local disks there. So there is the peak parallel disk system that one can use for input and output. But there are some cases where you really would like to have a higher bandwidth to the disk. And there are 40 nodes with this NVMe. You can think them as a kind of SSD like disks with 3.2 terabytes per each node. So if there is a case where you really need to write a lot and read a lot from disk during the simulation, this can actually give a much better performance. And for the GPU nodes, all these GPU nodes, they have the same local fast disks. Also there are many machine learning methods might need that kind of disk access. The CPUs in GPU part, they are just the same as in the other part of the system. And as I said, the GPUs, they are the NVIDIA V100. Each GPU has 32 gigabytes of memory. The interconnect between the nodes, which you need to transfer the data between nodes. And as Ivan already discussed, sometimes you can think as really the heart of the supercomputer in the sense that what really makes up typically supercomputers is that you have lots of CPUs, lots of CPU cores, and that they are put together via this fast interconnect. For the CPU part, the speed of this is independent. That's 100 gigabytes per second. And for GPU part, that's 200 gigabytes per second. So one question came up in the PuTTY section. Someone asked, as an alto student, are we supposed to use Triton or can you also try out PuTTY? And I guess it would depend on being both undergraduate student and student who's also a researcher. Yeah, I can actually, I have a one slide at the end that okay, when you would be using, in what kind of cases, you might be using CSC services, what kind of cases local services, and also what you need to get access to these services. So I think that will be a perfect question. I think for the purpose of this course, I guess you will be using, and I mean for the more hands-on part, I guess you'll be using mostly Triton. But I can already say that if you are familiar with using Triton or any similar cluster, then switching over to CSC should be relatively straightforward. The environment is very similar. As said, Mahdi, that's really meant for large-scale simulations. So in practice, you won't get, you won't have the possibility to run anything using the whole Mahdi, unless in very special circumstances. But in principle, the interconnect and system is designed so that in principle that would be possible. And I mean, that would mean using about 180,000 cores in total. So I guess it's number, which is easy to understand that this is actually quite huge. I mean, in my laptop, I have four cores, and here I have almost 50,000 times more cores. As said, they both are, or the CPUs, mean Mahdi, they are AMD CPUs. They run a bit higher base frequency, 2.6 GHz. There is some possibility to boost for that in simulation, and the amount of memory, as I said, that's heterogeneous, that constant per node. The GPU part, as I said, is at the NVIDIA 100. And the Infinity Band HDR in principle, it's very similar to Pufti, but the bandwidth over the whole machine is this higher 200. There is actually, it should be not gigabytes, but gigabits per bandwidth. The theoretical peak performance of how many floating point operations, additions, multiplications, and so on per second the machine can perform. That's 7.5 petaflops, so 7.5 and then 15 zeros. Maybe to mention at this point and so that in how to use these systems and if you want to find the more information, we have the docs.csc.dev. Which is the main user documentation. And if you go here under systems... Could you zoom in some? Sure. Can you zoom in some? Is it better now? Great, yes, thanks. Yeah, so I mean, you can find a bit more details about the CPUs, both in Pufti, in Mathti actually the details of the nodes and the AMD CPUs, they are actually quite complex. There is deep memory hierarchy there, so if you're interested, you can find more information about that here. Okay, there is... Maybe the answer in this technical part also, there was this concept of node that is clear, that was not clear to everybody. So, yeah, the way the supercomputers nowadays typically build is that of course you have a huge number of CPUs and each CPU they can have lots of CPU cores and then attached to the CPU you always have some memory, some physical memory, some DRAM modules and so on. And I guess the main distinction is that within a node all the CPUs, so for example if you have two CPUs in Pufti node both CPUs and all the cores within all CPUs they share the same memory. Let's say this is 190 about gigabytes of memory that we have in Pufti nodes. All this memory is accessible directly to all the CPU cores. So you can just read and write directly to this memory. And then if you want to preach the memory in the other nodes that cannot be done directly, so that will always go via the interconnect and even more so that you need to use some special communication protocols some special calls in your parallel program to actually access the memory in the other nodes. I hope that this becomes a bit clear, I mean if you start to use these systems but that's the main distinction, within a node everybody that sits within a node they can sort of share the memory while between the nodes something a bit more special needs to be done. About the cloud services, the main distinction between or let's say one benefit of using CC cloud services instead of directly using the supercomputers is that the operating systems that you have in the supercomputer they might be sometimes a bit special or at least they are not necessarily the most recent ones so stability is very important for these big supercomputers and also not always all the libraries that you need can be so easily installed there and there are definitely cases where you might want to have a bit more control yourself you might need to have admin rights and so on for setting up your simulations and that's of course not possible in supercomputers but then on the other hand that's something you can do with the cloud services of course you have also used it or let's say setting up business a bit more complicated so you need to administrate your own typically Linux system you create the virtual machine image and then you can sort of use that over the internet and as I said we have two main variants of these cloud services we have the C-POTA which is you can normally access over internet and A-POTA that's accessible only from special network so that's for the sensitive data and if you don't need the full control for the operating system etc nowadays using containers like docker or singularity or things like that is quite common so there is also practice service available for these kind of use cases and depending on what you actually want to do with your within the cloud you can get to different hardware flavors also at CSC so you might might get something which is more HPC oriented you might want something which has GPUs or something which has better IO bandwidth for the disk access maybe just briefly that okay if you very often you don't really need to do any programming if you're using the supercomputers whether it's CSC or Triton or any other computers in FGCI if you're just using some existing software you can in many cases use that without brilliant for parallel programming but if you need to do that there are multiple of programming languages supported FORTRAN is still, despite the long history, quite relevant in high performance computing scientific computing then of course C++ Python and R have become quite popular in recent years Julia is something and then for the parallel programming there are the message-passing and the shape-memory-programmed OpenMP they are the most most relevant ones for GPU programming in Mahdian Puhti there is OpenACC which is sort of maybe something a bit easier to do and then there is Therese Kuda nowadays actually also one can use this heap that Ivan mentioned already and also OpenMP for data analytics, machine learning, deep learning there are various frameworks, TensorFlow, PY towards Keras et cetera that are readily installed at CSC and you can use it such Okay, a few words about the forthcoming monster computer, Lummi European Commission had about two years ago a call for hosting and maintaining a tree really big supercomputers with aim-to-be Xscale is something, I mean, CSC current supercomputers as I said they have the performance of order of few petaflops so with 15 series rows there and now we are starting to approach the Xscale era where the performance is something like order of Xscales and the exaflops of 1 and 18 zeroes and there will be three of these pre-Xscale so not really yet the actual Xscale but something that they way forward that to be built in Europe and the consortium led by CSC was selected of one of these so this consortium consists of in addition to Finland, Belgium, Czech, Denmark, Estonia, Norway, Poland, Sweden and Switzerland and the machine that we will receive hopefully by the end of this year that will have about half Xscale performance so about 550 petaflops it will consist of performances mainly given by GPUs and this actually they will be now AMD GPUs for you probably the interesting thing is that 25% of the resources that's dedicated to the Finnish users that's based on the proportion of funding that Finland put there half of the machine is also available in principle any researchers in Europe so it's possible to apply for access that even outside the consortium members but Finnish researchers via CSC have a possibility to access with we don't know the exact process procedure yet how that will be carried out probably similar to current so-called grand challenge close to CSC resources maybe few things to mention here that at the same time it's both huge possibility but it's also challenge so normally supercomputers as any computers they get old, they date and the typical lifetime is 5-6 years and normally when we get the new supercomputer that's a couple of times more faster than the previous one for example the Mahdi supercomputer is about 4 times or the absolute performance is about 4 times that of our previous computer how now in case of Lumi it will actually be 30 times more performance and there is really big paradigm shift in the sense that at this point the GPU part in national supercomputers there have been only some smallest add-on there but now the main part is really for GPUs and in order to really sort of fully utilize that one also first of all needs to think big regarding scientific problems and there is also quite a lot of work for making the applications work efficiently on GPUs the way one would program for this computer somehow it's not that different so Fortran and C++ probably the most important language is still with some Python and RU seeds for parallel program between nodes still the similar MPI approach that used at the moment but then of course we have the program for the actual GPUs and especially the fact that NVIDIA has really been dominating the high performance computing market regarding GPUs for a long time and the software ecosystem has been quite mature there now with AMD there are a bunch of new software frameworks and program frameworks for program for these AMD GPUs so he's been principle it's a C-like extension to C program language which in principle should work similarly both in NVIDIA and AMD GPUs and then there is this bit high level program with OpenMP and of course many scientific libraries will take benefit from the GPUs there is a question what does pre-X excel mean? yeah well it means that it's something it's supposed to be similar to the machines that will reach the actual exascale performance exaplop performance but it doesn't yet have exactly that kind of power so in this case it means that it's something which is close to exascale one can think which very probably need to use same programming approaches that you would use for exascale something maybe mentioned still illustrate the scale of this machine is that I think if one tries to evaluate the power of Lumi in units of modern laptops I think the pile of these laptops I mean if you put them together that would be more than 10 kilometers in height Lumi on one hand that will be I think the physical size will be about 10 escorts filled with younger people don't necessarily know that what phone boot is but I think the single cabinet that is the basic building unit that is about the size of phone boot okay there was a lot of talk about supercomputers but one thing to keep in mind is that these supercomputers they might generate lots of data and some cases I mean even if you're not using the supercomputers I mean there is still always data related to research and the data management I think that has becoming more and more important maybe not relevant yet for the undergraduate studies and so on but for example when you are making a proposal for research grants typically you need to know what is included also some data management plan one often nowadays speaks about so-called fair data so findable accessible interoperable reusable and idea is really that you try to describe to data so that it's for others it's easier to find accessible of course means that you put that data available in some public place and that it has a well-defined format so that others can use that and so that it can be used also later on and CSC has also various services related to this managing data in different parts of its life cycle I already mentioned the alas storage so that one can think for that mainly meant for so-called active data so something that is still part of the ongoing research project and alas as such it's more like just the storage service where you can store data either from CSC supercomputers or from your university workstation or from your own laptop and you get access to this alas service similarly than to the CSC supercomputers I mentioned previously that using the supercomputers as such is free for researchers in most cases however there is a concept called billing unit so which you can think of CSC money or something like that I mean you need to apply for some compute resources and the resources will be granted in units of billing units you can find a calculator on the pages how this billing unit actually goes an important point here is that storing the data in alas it also sort of consumes this billing unit an idea of billing units is of course even though using the storage or the supercomputers doesn't cost any money to the users of course costs some money to CSC and ministry so aim is to try to motivate and make sure that the users are also using that efficiently so you see we have a few more minutes before we need to go to the break okay so then I'll try to speed up here Pete so there are two basic data management services that you can access via CSC one is sort of more European wide EU. and there are a couple of different services some of them are similar to let's say Dropbox or Google Drive and so on but they can be also customized for needs and these are good if you have personal use it so you need to share data between international collaborators and then there are the sort of more national services which have various tools for storing the data but also providing metadata describing the data you have there and also tools for searching for the data these services you typically have to apply separately to get access there and some of these services they actually operated via the local contact person at at your university if you look in the CSC user documentation there is the sort of data part here where you can find some bit more information about these kind of services and here in accounts you can also find a bit information about also these data services and this is just overview of the different components so EDA is the basic storage service, KUVINE is for describing and ETSIN is for finding the data part creepy about the support services CSC maintains quite large collection of different scientific software so once again we can go to DOCS and here applications you can need to look them in alphabetical order what we have you can look by discipline so biosciences, chemistry and also that are they available in Mahdi or Puhti Puhti has a larger collection of software and Mahdi bit smaller one but you can find here that which software are available in which machine you can get from CSC you can get help also in how to use some of this software, sometimes also in planning states that if we have some idea that which software would be most suitable for the project of course in many cases the researchers themselves they know the discipline well but CSC has also some expertise in this area so many of the people who work they have background in science and have been doing a PhD postdoc and so on and have also made the tools and methods one service where I am particularly more involved is that if you need help in improving your own software so that it would run faster in our supercomputers or scale better in parallel and so on we can provide help also in that the main that this service as well for other ones is the Service Desk at csc.fi we run quite large amount of training every year related to different aspects of using our supercomputers doing parallel programming sometimes also how to use some particular scientific software and also for other services that csc is is providing these are typically two or three day short courses at the moment mostly in online format in normal times mostly at csc probably in future we provide more also streaming possibilities for them and depending on bit course they are either free of charge or have a small typically 40 to 50 years per day fee most of the material for trainings we develop they are available via github or in our training portal with some open source license okay so how to actually get access to csc services in principle if you have Haka user ID as most of you have you can just go to my.csc.fi and get the personal csc account so you cannot directly use your Haka account but with the Haka account you can get the csc user account in order to really get these billing units there are some sort of prerequisites so it's not everybody can apply for so basically there needs to be a project and a project supervisor and supervisor typically needs to have some experience so basically it doesn't need to be professional so postdoc level is typically enough and the project supervisor can then add other people in principle any people that has this csc user account can add to the project so for summer studies for example in principle your summer assignment supervisor they apply for the resources services billing units and so on and then they can just add you to the project and after that you can connect to csc supercomputers and use the supercomputers okay and I promised as a sort of summarizing also that when to use csc services if you look the actual individual course or GPUs at supercomputers they are necessarily not that much different than what you have in your local laptop and definitely not different than what you have in the sort of more local clusters for example the individual CPU core in my own laptop is a bit faster than the one in puhti so the added value of course comes that instead of using four cores in my laptop I can use 400 in puhti also it might be that there is some software that is available or it's easy to use csc supercomputers is that installed there and of course memory typically you have only a few gigs of memory in your local system in university cluster in fgci cluster you might have actually also quite large amount of memory there but in csc definitely there is at least this 1.5 terabytes maximum you can use and when you really need to do large scale parallel computing then you don't necessarily have enough local resources I guess you will be playing around with the module system with the slurmpatch system during this course and as I said it's very similar to what we have at csc so moving from the return to for example csc it would be relatively easy and as a final thing still here are the main pointers so the csc main webpage there is some sort of more general information to researchers at research.csc.defi and then the sort of main user documentation I already showed you with that thanks you for your attention and I don't know do we have still some questions or do we have time for a couple more questions yeah I think most of the well the most interesting question is to me is csc services good for or csc services a good solution for long term data storage as in 30 years or more csc provides some really long term preservation service so there is this kind of service it's not part of this normal so there is typically special agreement needed for that and that really is a long term preservation service and one aspect there is of course that the data formats and so on I mean they might be completely changing and these long term preservation services they care of that also what we are actually missing at the moment is and that's basically because how ministry sees our task is that for this midterm so something let's say between 5 and 10 years we in principle we don't have a service so for example Allah sees meant to in principle a bit shorter term service and as say this long term preservation that's then a bit more complex so something like 10 years or so that's something that we do not have at the moment but if you go to this fairydata.fi you can find more information about also this long term preservation great so yeah I guess you can continue reading the HackMD and answering like following up any further questions there yeah I can do that for a while there were a lot of questions about how the cluster actually arranged like what's a cluster what's a node how do I run my code on that this is the kind of thing that will get into great detail on Tuesday and Wednesday so if it's not clear right now don't worry like we're slowly getting up to that point and you will see it then so with that said I guess we will take a break now for