This post is a reflection on the DeIC conference I attended in November 2012.
The conference marked the initiation of the new organization, DeIC and I think for everyone, including me, it was quite a brainstorming event – trying to get to grips with the what this new organization is supposed to do.
I had invited Vangelis from our Greek sister NREN, GRNet. He’s managing the Okeanos project, which IMO has already achieved much of what DeIC aims for. He gave a well-attended talk in a parallel session and also participated in the panel debate “The future of cluster computing – in the basement or in the cloud”. As seen in the picture above, the turn-up was quite low – unfortunately there were parallel sessions that drew more interest. In other words, despite good intentions, we did no succeed in producing an animated and broad debate on the subject. IMO we need such a debate though – preferably with real users and stakeholders.
The sessions included inspirational talks from both our CEO and head of board – both repeatedly stressing the official subject-line of the conference, “Data, computing and network – in the service of research”, and the importance of three new main assignments from the board, to ensure the availability of compute and storage infrastructure services:
and to make sure that such services are actually used by researchers – preferably researchers who did not previously use computing in their research:
A bit of background information to understand the challenges of this:
DeIC is a merger of the former Danish funding agency for HPC equipment (Danish Center for Scientific Computing – DCSC) and the Danish NREN (Forskningsnettet). The merger is asymmetric w.r.t. head count: Forskningsnettet had 40+ employees, DCSC had one administrative head and indirectly funded 5 HPC sysadmins. Budgetwise, there’s more symmetry.
Previously, DCSC awarded funding directly to researchers, who then simply “sent the check on” to their local HPC sysadmin who carried our the actual hardware purchase, mounted the hardware and set it up with operating system, job queuing system, shared file-system. He, or they, also provided support for running jobs on said queuing system, storing data on said file-system and installed any needed software.
This is how “delivery of e-infrastructures” was and is still carried out – perhaps servicing some 50 researchers* – all from the natural sciences. The board wants to change this: increase the number of serviced researchers and extend the use of e-infrastructure beyond the natural sciences.
The question is how?
Readers of this blog may be aware of my opinion on this: We (DeIC) should:
- build a scalable virtual machine provisioning service
- build a scalable, distributed HTTP file store
- build a web-based self-service interface to this
- build high-level services on top of this – data processing services
- build worked-through examples of how to use the services in each scientific domain
There’s a whole lotta building in the above. This is my view – pretty unchanged since 2009. A number of building blocks have emerged since then, but the fact remains, that it is a very big challenge and there’s still no off-the-shelf solution out there.
In coming posts I’ll describe how I propose to go about the building. I stress that this represents my personal view. How things will actually go will depend on external factors – like input from actual future user communities. Anyone interested in this are welcome to make their voice heard via the DeIC website.