The ETP4HPC Workshop at ISC'17 in Frankfurt, Germany allowed us to talk with Michael Malms, who is working at the ETP4HPC Office. His responsibility is to bring together strategic documents like the multi-annual roadmap (SRA), orchestrate workshops, and also pushing certain ideas like the extreme-scale demonstrator idea that the ETP4HPC Workshop focused on. Michael Malms has 30 years of experience in systems development. He worked for IBM before in the mainframe development arena. He then moved on to HPC Systems. He led a team working on blades for the Supercomputing Center in Barcelona in 2004. This team also delivered the compute node for Roadrunner in 2008. So Michael Malms has some experience in this area.
When ETP4HPC was founded in 2011, he was representing IBM in the first Steering Board. Later on, when he retired from this part of his life, he joined the Office as a supporter for any kind of strategic definition of roadmaps, programmes and so forth. ETP4HPC is a consultancy body to the European Commission. ETP4HPC proposes certain topics and certain areas. What ETP4HPC proposes is, and what Michael Malms' main responsibility is, is to bring this information together, not just by himself but with colleagues from ETP4HPC. There is a nice set-up of working groups and working group leaders. All the members of ETP4HPC have a chance in contributing to this roadmap, which is published. Michael Malms showed one of the latest issues of the SRA roadmap.
Currently, ETP4HPC is working on the next issue, making sure that most of what the members say, really comes to fruition and will be part of the next work programme. This is what Michael Malms is very much interested in and what he is pushing.
The production process of the roadmap is the following. ETP4HPC basically is putting out one of these documents each time a new work programme comes out. Now, for instance, the 2018-2020 Work Programme comes out. This will be published in October 2017. ETP4HPC likes to issue a new release with an updated roadmap, and technology areas, just a month before that. So, all the groups who have an interest to put in a suggested project can refer to this document. They can say: "We read this and in our proposed project we refer to these milestones in this chapter, and that milestone in another chapter. This is what we propose."
The process is that ETP4HPC has founded eight working groups in 2012. ETP4HPC is still using these eight working groups. They are organized in a kind of specific fashion. There are three areas like architecture, system software, mathematics and algorithms in one dimension. Then there are the brain compute subsystems and efficiency in energy and system resiliency. They are organized in an orthogonal fashion because these two things really touch on all the levels of the HPC stack. ETP4HPC also focuses on new compute models. This is particularly interesting now. ETP4HPC is not only focusing on traditional HPC technical computing. It becomes more and more obvious that Big Data, HPDA, Artificial Intelligence, and Deep Learning are in need of the right infrastructure. ETP4HPC has to learn from these areas and they hopefully learn from ETP4HPC. HPC in a few years will be totally different. This is what ETP4HPC has to update in this document, each time it has a new release.
What has been new in the last three years? ETP4HPC collectively in the working groups has thought about this. It will be put in the document and reviewed by all the members. The ETP4HPC members all have a chance to contribute and to say: "I see this a little different." Of course, in the end, somebody must say: "We need to follow a certain path. We cannot make everybody happy." But so far, so good. This is now the third release that ETP4HPC is working on. It worked pretty well, so far.
It is an editorial process. ETP4HPC has an editorial team. Michael Malms has two colleagues who work with him. At the end, they have to stand for the content. There are, of course, multiple reviews with the working group leaders first, and with all the working groups afterwards. There are about 150 people organized in these eight working groups. It is quite a good audience.
There are 80 member organisations from large companies to small companies, from research organisations to universities. We asked how the balance was. Research, of course, will have an important say, but how is the balance between the small and the big companies?
Michael Malms answered that there are quite of lot of SMEs. There is a special dedication to their needs. In the next version of the roadmap, there will be ten critical points that are crucial for SMEs to become successful and have more success and more impact in the SME provider's side. In ETP4HPC the SME technology provider side is represented, and not the SME users. The SME providers have their issues and their challenges. ETP4HPC would like to support them more. One could say that everyone who is doing HPC in Europe in terms of R&D is welcome - because this is the criterion - ETP4HPC has only members in the technology arena who do R&D. It is not sufficient to only sell computers and HPC. They have to produce it and have an engineering team on board, whether this is ten people or 2000 people, that doesn't really matter.
What ETP4HPC doesn't have at this point in time - and this is why it is organizing the workshop - are industrial users. They could join as a member, they could participate by communicating their requirements, very much like HPC centres and Centres of Excellence give their ideas and their contributions. This is something for ETP4HPC to work on. Michael Malms hoped that with this workshop, ETP4HPC would get a better impact here.
ETP4HPC is very much bound to Horizon2020. ETP4HPC consults the European Commission. Most of its initiatives go into the FET area. Some parts are going into the infrastructure arena, like the Centres of Excellence but this is all around Horizon2020. When one approaches more the end of Horizon2020, in two years, Michael Malms assumes that ETP4HPC is going to be asked for its contributions and ideas for the next Framework Programme. He hopes that ETP4HPC will not end with Horizon2020 but will just continue on and maybe joining forces with other ETPs. ETP4HPC has now a close relationship with BDVA, the Big Data Value Association.
ETP4HPC is going to have a workshop pretty soon to discuss other options for very close
collaboration and to discuss its vision for the time within five years when it may be doing more together, maybe in one platform. That all needs to be discussed.
There is also this new initiative, called EuroHPC which came about a few months ago for which the Member States say that they want to invest more and do more about HPC. We wanted to know whether there is a role for ETP4HPC in this.
Michael Malms replied that this is mainly a procurement activity, to his opinion at least. On the academic side, there was not any real interaction so far. ETP4HPC understands that the governance structure will be defined by the end of this year. Late in the Fall, ETP4HPC is going to understand who is who and what one wants to do in terms of organizing the procurement of these machines. Michael Malms hoped that ETP4HPC can propose certain directions of what these machines should look like. There is also another initiative called IPCEI. This is the same story. There is a difficulty. There are multiple players on the European and on the national side. Michael Malms' personal concern was that there is not any overriding governance nor a strategic organisation fitting this all together. Everybody is working for themselves with their scope, their intention but, of course, at some point in time, one needs to organize this.
The workshop is concentrating on demonstrators. We wanted Michael Malms to tell something about it.
This idea came up in 2014. ETP4HPC had a look at the projects launched at that time. There are 20 research projects. By now, there are 25 already. One can assume that maybe in a year or two, there are 30 projects. They all have their results and small prototypes, whatever they do. Then, it is assumed this is it. At the end, you have maybe 40 or 50 results, not tangible and not really synchronized. ETP4HPC thought that this is not good enough. One needs to organize something that is soliciting a combination or an integration of several of these technologies. ETP4HPC suggested HPC prototype systems to be integrated and used. ETP4HPC calls them extreme-scale demonstrators. ETP4HPC now pushes the provider side, the HPC Centre side, the Centres of Excellence and the industrial users to understand what this concept is and to take advantage of it. Since 2014, there is now the European Cloud Initiative that came in with two pre-exascale machines being procured, two exascale machines being procured by 2022. One might have a roadmap even between these extreme-scale demonstrators as a first step, and pre-exascale and exascale machines on a commercial level. This is to prepare Europe for tangible HPC products; real, entire, complete systems. Michael Malms hoped this will be a success.