Today we had presentation with Lucas to defend the results of the EDGE project, which is the CPER project within the MISN CPER funding a large part of the Grid'5000 node in Nancy. Those who are not used to the complexities of the wonderful academic funding world in France, consider the CPER projects as "joint ventures" between entities such as the research ministry, the regional council, the INRIA institute and the European council (in our case), and on a specific "market share" being scientific projects.

The EDGE project is on Experimentation and Distributed Systems at Large Scale (the acronym is in French). All the details can be found in the project document (in French). Of course, the slides below, should be instructive too.

Download PDF

The format were 20mn of presentation and 20mn of questions and interactions with the experts. Since my presentation was quite short, we had a very interesting discussion session with our three experts. Here are the questions that I remember about.

Vincent Breton first asked some details about the study of the gLite middleware on top of Grid'5000. He wanted to know when the preliminary technical work will be over, allowing the actual scientific assessment. We hope that it will be before fall, since things progress nicely, even if not as quickly as we'd like. @Sébastien Badia, do you feel all that pressure on your shoulders to hack it done? (hehe, I'm clearly addict to social networks these days)

He also asked about ScaLab, our project to get some coherence between the differing experimental methodologies. During the discussion, we specified that our workflows are a bit different from the ones we usually think about on grids: they are control-flow processes, not "classical" data flows. That is why we contacted Francois Charoy since that's his expertise field. He concluded that this is an interesting approach, which is really too bad since the ANR project that Lucas built on this got rejected recently...

Patrick Amestoy wanted to know what GPGPU will change in our settings, whether we will buy such machines, why doing so in Nancy, and the impact it will have on the SimGrid framework (yeah, that's the way it goes in such circumstancies: you get 4 questions in one, and that's completely ok: I do the same when I ask questions ;) ). GPGPU will naturally change a lot for us, we really need to buy some such machines in Grid'5000@Nancy. The main motivation is economical: GPGPU are much more efficient and less greedy, which is unavoidable when you know the power and cooling costs in data centers (that's the point I tried to make in this presentation). As a result, everybody will move to such systems (when possible, not all problems fit to these systems) in the future, so we need to acquire some hardware to tests it. Why in Nancy? well, because there is several kind of such GPGPU, and the mutualization resulting of Grid'5000 maximizes the diversity that each of us can access to.

Simulating GPGPU within SimGrid is a really complex issue. We will definitely tackle it, for example in the ANR project following USS-SimGrid that we wrote this spring (more on this when we get the evaluation result, in July). Even if the ANR gets rejected, I wouldn't be surprised if Arnaud would try to address the issue anyway. It will be extremely hard, but we'll strive to develop models of these systems for SimGrid, and assess their validity on Grid'5000.

Pierre Kuonen asked a lot about the software tools we're building (simulator and emulators). He wanted to know whether they can be used separately or whether they are bind anyhow to Grid'5000. Since there is no such strong link, I think that he may consider using them in his own research, where it's hard for him to get access. I'd be glad! He also asked when it will be ready to use, and that's a bit more difficult: we need 6 other months, at least.

Much discussion were also raised with the experts about our difficulties to create some contacts with the researchers of the industry in our region. We were given several pointers, and I hope that we will manage to get something flowing soon.

At the end of the day, the overall comment from the expert were that our project is now much more mature than last year. They had a few recommendations anyway. First, we should continue organizing our community. A proposition was to organize a user seminar, gathering effective and potential users, along with people working on this. Then, they asked for a better cartography of what we are proposing. It seems to me that a presentation by use case would be easier to understand than our current presentation by approach and methodology. At least, that's what I understood from the oral debriefing, and we'll have to wait for the written feedback any time soon for more details.

That's it. It was a quite intense day, with several interesting propositions to improve our work. I'm thankful for our experts who came from the very far away just to help us improving our workflow. And of course, the fact that they have no strong criticism put my mind at ease, too.