jump to navigation

Funding Equation for University Computing June 3, 2008

Posted by gordonwatts in computers, physics.

Racks full of computersIt used to be that when you wanted some computing at your university you got the computers and then setup a room somewhere, plugged the network, hired some undergraduate to manage the things and you were off to the races.

That was then. It is different now, and it is getting more different as time goes on.

Several factors are coming together. Perhaps in University research the major driving forces is the overwhelming need for CPU cycles. Those cycles are no longer cheap, either. When you need a lot of computing power you can’t just buy that 500 buck computer off the shelf. You have to buy the 1U pizza boxes, stack them into special racks, run special power feeds, special room cooling, etc. This power and heat density is another driving factor. To do this you need a room where you can install all of this – with the power and the air conditioning. Very few buildings not designed for those sorts of loads have it – and retro-fitting the buildings – even ones just 30 years old – is amazingly expensive. Your other choice is to buy servers at a lights-out facility (a data-center). No matter how you do it, it cost money. A lot more than it used to.

This is affecting how we do computing at the University. You can no longer say “ok, we’ll give you 1 million dollars of CPUs in your startup fund” to attract new faculty. You also have to add “and the necessary $700K in order to install, power, cool, and network the computers”. I get that 2/3rds from some recent discussions with people installing large amounts of computing. This is making it harder for universities to make attractive computing offers in their start-up packages. I don’t think anyone has blinked yet (the university usually swallows the cost somehow – but someone – the department? – has to pay for it in the end).

The other place this affects things is in grant applications. I’m not so familiar with the DOE, but at the NSF they always look for an in-kind contribution from the host institution with the grant proposal. Frequent agency incentives are things like reduced rate shop time to make a proposed detector. When it comes to computing it is things like supplying the infrastructure to do the computing. But as costs go up and up Universities – especially ones that don’t have super-large endowments – are having a harder and harder time not passing the buck. The question is which university will blink first and refuse to support the complete cost? And how quickly will the others follow on. And what that will mean for future compute funding for research groups.

Or perhaps a change in the way we host computing?


1. David Pace - June 3, 2008

I know you are talking about large scale computing, but there are some cases in which simulation and modeling has become much easier for individuals to perform. There are many plasma physics codes that simulate turbulence and related phenomena that can now be run by individual experimentalists. Such codes require a multi-core computer, but those are now “cheaply” available to anyone, albeit a 4 or 8 core variety. Of course, this only allows the experimentalist to perform the linear calculations because the full nonlinear runs still require a full cluster of the type you discuss above. Even the linear runs can be very useful for comparisons and the ease of use for the codes is encouraging such work. The codes are becoming more like any other tool and they can be applied more frequently to experiments.

This is leading to more comparison between measurements and theory in the field of plasma turbulence and transport. Such comparisons are the very basis of physics research, so this is a fantastic development. I can’t imagine what physics research will look like if (when?) machines with thousands of cores become available to individuals.

2. telfer cronos - May 6, 2012

“your’re bullshitting me”. Computing has never been cheaper. In the 1960’s a few companies thought it worthwhile to spend millions of dollars on a computer that was much less powerful than today’s movile phone.. They had airconditioning, false floors for cabling, and lots of operators and technicians to keep them running and keep work up to them

3. gordonwatts - August 6, 2012

Hmmm… looks like I wasn’t clear – definitely, it is cheaper now than it used to be back in the bygone days. But compared to just a few years ago, I have to spend quite a bit more per box when I get them installed in the university. And David, I hope that the changing architectures of computers continue to lend themselves to more and more different physics problems with greater efficiency. particle physics has already gone through that (perhaps it will again??) and just lots of cores with about 2-3 GB/core is all we need.🙂

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: