Virginia Tech on Your Mac Life

YourMacLife writes "On tonight's Your Mac Life, the Dean of Virginia Tech's College of Engineers, Hassan Aref, will talk about the G5 cluster the college is building and what it means to supercomputing. Questions can be sent in advance to onair@yourmaclife.com." See the web site for more details.

Comments

Re:Did BSD make this possible?

[Juanvaldes]

Does this mean they're planning on running some variant of BSD? I would imaging that, for licensing sake, they wouldn't put a stock Mac OS on there (OSX)... would cost "too much" and would provide "more" than they need.
Unlimited Client X server costs only 1000.

Re:Did BSD make this possible?

[The Bod]

Does this mean they're planning on running some variant of BSD?

Yeah, it's called MacOS X. (early version of Panther)

I would imaging that, for licensing sake, they wouldn't put a stock Mac OS on there (OSX)

What? They are buying 1100 machines, they get 1100 copies of MacOS X. What kind of licensing issue are you dreaming of?

Re:$5m??!

[adam872]

Compared to the ASCI series put together by the US government, $5M is not a lot of money. Consider that a single Sun SF15k, IBM p or z series or HP SuperDome can cost easily this much (that's just one machine).

I think VT are getting a pretty good deal. Really large clusters/supercomputers *can* cost upwards of $30M, depending on the configuration.

Notes From the Interview

[n8_f]

Well, I browsed through the links at threshold=1 and I couldn't find any comments from anyone who had actually listened to the broadcast in question. No surprise there, I guess.

It wasn't that great, so you didn't miss much. It starts at 1:17 and ends at about 1:37 in the archive file for those who would like to listen. For everyone else, here is a list of the highlights:

They didn't go with an Xserve chassis because they couldn't get one with a G5.

They are using the stock chassis, no modifications.

It will take up about 6 to 8 racks. (Which seems really small to me; they must be packing them in tight).

Using OS X for now, the stock install (10.2.7). They have had other people approach them and they will consider other OSes in the future, but they are going with an OS X install for now (this seems to be a lack-of-support issue).

They do have a 400-node cluster running now, so they aren't complete novices to building clusters. But this was still very new for them and several times Mr. Aref mentioned that Apple had helped them out a lot, going to some of the other vendors involved and talking with them on VT's behalf, etc. Apple obviously sees this as a very strategic project for them.

They chose 1100 nodes because that hit the sweet spot for budgeting and where they wanted to be on the TOP500 list.

The cluster comprises over 19 tons of gear.

It will be a node on a network of supercomputers nation-wide - the National Lambda Rail initiative.

They (VT) are creating a video documentary of the whole project that it sounds like they will put online later in the year. They are also writing a handbook as a kind of howto for building your own cluster. Sound very willing to talk to other groups about how they did it and help others along.

That is about it. Not much as far as technical details. Mr. Aref said they will release all of the technical details later. He wouldn't venture a guess on where they will be on TOP500 until he's seen some benchmarks, but they obviously expect to make the top 10.
Personally, I am extremely curious about this whole project. Using a desktop chassis seems like an unconvential way to build a large cluster, so the photos will be very interesting. But the documentary! I think that is awesome. This might provide a unique perspective into how a large cluster is built (Mr. Aref joked that it involved a lot of pizza).