Weta Digital Supercomputer For Hire

sushi writes "NZ's Stuff news site is reporting: 'Peter Jackson's special effects shop Weta Digital has teamed ... to establish a world-class supercomputing facility in Wellington which will be rented out to clients worldwide.' Currently comprising 504 IBM blade servers, each of which contains two 2.8 Gigahertz Intel Xeon processors, 6 Gigabytes of memory and 40 Gigabytes of storage, and ranked 80th in the top 500 supercomputers, they are intending to upgrade into the top 10. Also covered at the Australian Financial Review."

Comments

Dupe?

[Anonymous Coward]

Rent a Bit of Weta Digital

Re:Dupe?

[wrinkledshirt]

Maybe Slashdot should rent the machine to keep track of stories with similar content...

top 500 !

[phreakv6]

This is the top 500

Correction?

[iamdrscience]

Hmm... if you look at the top 500 it appears that 80th was their previous place and that they have since upgraded their cluster to become 77th.

Doomsday scenario

[ciurana]

Hrm...

How soon before the bad guys set up a dummy corporation and start running nuclear bomb or protein folding simulations on this cluster? I'd be very interested (probably along with some governments) in Weta's and Gen-i screening processes. Will anyone who can foot the bill get access?

I know, this is tinfoil hat stuff, but it's late and I get this "glass half full" visions when I'm sleep deprived.

Cheers,

E

Re:Doomsday scenario

[Vellmont]

How soon before the bad guys set up a dummy corporation and start running nuclear bomb or protein folding simulations on this cluster?


The hardest part by far in making a nuclear weapon is getting the fissile material. If you are able to get highly enriched uranium you don't even need to do any simulations, the design is fairly simple and no testing is needed. Plutonium is a bit harder.

The point though is that computer simulations of nuclear weapons is the least of your problems, and is by no means required. Computers aren't secrets, and getting a few hundred of them together in a cluster is a task anyone with $100,000 can easily accomplish. Compared to getting the required fissile material, any required computations are easy.

I'm not sure what you're getting at with protein folding. Is their some doomsday weapon you can create by knowing how proteins fold? Even if it is, it's not a big concern. No one has gotten even close to completely simulating a protein folding. There's simply not enough computing power yet. What's been done to date are just small scale simulations.

Finally!

[xsupergr0verx]

Thanks to WETA, now I can run Doom 3!

rent it out

[OneArmedMan]

to other ppl that need to render stuff. Im sure they could figure out some reasonable pricing vs CPU time etc.

An idea...

[Veridium]

You know, that only has ~500 nodes right? How many people out there are either out of work, or sick of doing what they're doing? Maybe we ought to get about 1000 of us nerds together in some kind of co-op, cluster our machines then rent it out? My main box is dual opterons and I already have 6 dual P-pro 200s clustered...

Yeah, I know, the logistics of it, the devil would be in every detail... Neat to think about though.

Re:An idea...

[Vellmont]

You know, that only has ~500 nodes right?


It has 500 nodes that are highly and quickly interconnected. It's like the difference between 500 people working on a problem in the same room, and 500 people spread across the country communicating by postal mail. Most interesting problems require a lot of inter-communication, so 500 slowly connect nodes isn't too usefull.

Re:6gig of memory?

[Gopal.V]

> almost 3TB total

Do remember that 90% of the time, it's not the size that matters , but the organization.

I've worked on a relatively small cluster processing experiment in college with 12 boxes on a 10 Mbps LAN with a combined memory of 1.5 Gb RAM . It might not look much , but with 32 MB of RAM on each box (each had 128 MB ram) being held by the home cooked shared memory daemon (this was waaay before memcached was born, Ok) , the boxes ran the number crunching beautifully .

The operation needed was simple, to sort and process an amazingly HUGE chunk of data in almost realtime (in this case some wierd algorithm some Mech teacher wanted and did up in C).

Anyway in about 7 weeks and reusing a dozen of the college's vanilla PCs we did a LOT of interesting things .

So my question is , how's the server connected memory wise (most of these tasks are highly memory bound or at least that's the major bottleneck to optimise).

Re:This is great

[Anonymous Coward]

If someone paid to rent a supercomputer for a few hours to break your SHA1 SSL keys, you've got bigger problems than just your keys being broken :)

Re:6gig of memory?

[zcat_NZ]

Yep. Enough for Doom3, or Longhorn. Possibly even enough to run doom3 IN longhorn!!

Re:Awesome

[Jerry Talton]

It really cracks me up that you rate finding large mersenne primes in the same category as curing disease or discovering other intelligent life forms. Talk about something with no practical applications...

Weta's old cluster

[SlashdotMeNow]

We actually bought the old cluster Jackson used (for next to nothing I might add!) It's 62 PCs (no they run Windows) totalling 124 processors (2 racks full)

That's a total of 124GHz and 124GB of RAM. We're using it to render architectural fly-through movies.

Hmm... I'm all hot now... Need a cold shower!

Noooooooo.... laptop hard drives...

[ricky-road-flats]

We've had big problems using IBM blades, not least because by default they come with crappy 5400rpm laptop hard drives - and the 40 GB mentioned in these blades imply that's what they've got (which is what my predecessor ordered).

I guess with 6 GB RAM each they shouldn't have to do much (or ANY, if I was running this) swapping, and if the jobs are tweaked to not use the hard drives too intensively, they might be OK. If what you do uses the hard drives for much, they are sh*t, to put it mildly. If you could plug these into the blades, they's be very useful, quick machines. But you can't yet.

The really crap thing is, if you do want SCSI drives in the IBM blades, you connect a module ot the side of the blade which gives you a couple of proper SCSI drive bays. Which halves the number of blades-per-bladecenter to 7.

Given the bladecenter is 7U tall, you'd be better off with 7 1U servers with SCSI bays already in and better NIC options. The internal networking of the bladecenter is awful for everything but the simplest low-requirement setups - it's hideously expensive to give each blade a couple of gigabit connections.

Even these cheap little things are 1U, take 2 U320 SCSI drives, and have dual Gigabit connections built-in.

And I *still* can't get USB dongles to work with thes fscking blades, grumble grumble.

Having said all that, when can I play on this thing? My Folding@Home could do with a bit of a boost, and with Hyperthreading I could have 2016 units running simultaneously.... although it might get a little warm behind the racks, 1008 2.8 GHz Xeons pump out a good bit of heat!

Connectivity Question

[dosun88888]

Do they use Tolkien Ring?

~D