Stanford Gets First Sun Blackbox

miller60 writes "The Stanford Linear Accelerator Center (SLAC) will be the first end-user to get a Project Blackbox portable data center from Sun Microsystems. The 20-foot shipping container (which will be white, not black) will sit on a concrete pad behind the computer building with hookups to power, a 10-gigabit network connection and a chiller located on an adjacent pad. The 'data center in a box' will allow the SLAC to expand its computing capacity even though its existing data center has maxed out its power and cooling."

Comments

Very impressive device

[nbvb]

Project Blackbox is one incredibly cool device. Sun was gracious enough to park one as a demo at my company, and it's just a very well engineered, game-changing design. The beauty is that it can be done relatively cheap, because shipping containers are CHEAP in the US. Most of them come from China, and since we import more than we export, we're stuck with a boatload (literally) of excess containers.

Imagine - rather than spending many millions building a true data center, you can just purchase a (relatively) cheap warehouse and line these things up inside. Instant data center - with lots of inherent redundancy.

Mirror one Blackbox to another across the warehouse.

Disaster Recovery? This the best thing since sliced bread. Park one at another facility 50 miles away and off you go.

I'm highly impressed. It's a bit cramped in there, but if you do your work neatly and place the servers in the racks correctly, it's not an issue. One shouldn't spend much time in the data center anyway!

and I asked ...

#1 - yes, they are standard racks, so other vendors' equipment will fit.
#2 - I asked about "oversized" equipment (such as Superdomes, E25k's, disk arrays, etc.) - they're working on a solution for that too. My guess is that it would involve removing some of the racks to make room.

I think Blackbox is a great idea with lots of deployment potential. Another thing to note - I was told that the air filters are designed to filter out lots of particulate matter -- sand included. You can guess why.

''A chiller system will be added in August''

[Alain Williams]

Oh, so how will they keep it cool until then ? Not switch it on perhaps ?

Re:''A chiller system will be added in August''

[davef139]

If you ever read up on shipping containers which is interesting stuff. Some are build for blast freezing in compartments of ships for like off shore fishing. I belive they can go down to -60C

Node Failure?

[tansey]

I'm just curious, but is the inside of this thing roomy enough for a person to easily get in there and replace a part? I know that node failures happen on a very regular basis with clusters and the box doesn't look very wide.

Stanford is a very appropriate place...

[foxtrot]

for Sun (Whose name came from where their first machines were seen, the Stanford University Network) to deploy their first of a new idea.

I'm not sure it's the world-killer that everyone wants to think, mind: If your data center is tapped out for power or cooling, you'll still need to get portable power and cooling to go next to your portable data center, but it does seem to be an excellent idea to tide you over until your real data center expansion gets built. Which means I expect to see a number of these sitting outside fixed data center locations in a basically permanent role, just like the "temporary" trailer classroom buildings outside schools and all the other stop-gap measures we implement "just to tide us over" that wind up being permanent emplacements.

I kinda fear this outside our data center. Especially when the machines therein get on the "long in the tooth" side, and we've decommissioned every application in the thing but one.

It's a great new idea, don't get me wrong, but the problem is how most companies want to run their data centers doesn't look a whole lot like how anybody's actually doing so in the real world. :)

They've got a serious need for more space...

[Anonymous Coward]

SLAC is in a big bind for space to house computers. They're out of power in their existing building but are also under very very onerous rules about approval for electrical installs due to an accident on site that nearly killed an electrician. They figure 24 months to get a new electrical feed from the transformers just outside the building... Apparently they'll be able to get the (simpler) electrical install of the Blackbox done more quickly.

So, they are using the Blackbox in the mode of "gotta get more capacity yesterday" vs. a real change in direction of datacenter planning... Still, I bet SUN sells more of these to customers in similar situations.

Re:the obligatory...

[Yvanhoe]

Sun can :
https://photos.sun.com/asset/7557?returnPage=/page /1183

old news in oil biz

[Jeek Elemental]

I used to design modules very much like this for the oil biz, its a common way to provide office space, utilities, subsea control, temporary functions etc. for oil rigs and hazardous areas.
You can order the shells pretty much any size/shape you want, especially if you dont have to worry about regulations.

Theyre not as sturdy as you might think tho, they get beat up something awful by transport, esp. offshore.
Crane operators like to use the one in the sling to knock the others in place/out of the way :)

So if you look at this but need some other function/more room etc. there is already a huge number of solutions and alot of companies who can tailor them or rent you one.

Eight racks, 250 RUs per BlackBox

[wwwillem]

A BlackBox has 8 racks (not 6). One of the racks is used for infrastructure components, like the dehumidifier, power, network, etc. The remaining seven racks are 38 RU, but because of the power distribution unit and a patch-panel, you can fit 36 1U servers in a rack. That is a total of 252 1U servers per BlackBox.

Someone calculated that if you would fill it completely with for example X2200 servers (two dual-core AMD Opteron), it would end up around position 200 in the Supercomputing Top500.

Re:My conerns...

[mollymoo]

I'm assuming you don't set this up 'in a parking lot' but under some sort of cover/tarp/tent, even painting it white, putting it outside in the northern california sun, can't be very efficient as far as cooling is concerned.

I suspect this is a lot like worrying about aerodynamic drag on a 100 ton locomotive. Yes, it exists, but it's negligible compared to the other load you have to deal with.

I suspected the solar heat load was relatively small, but I decided to run a (very) rough estimate to get a better idea of the magnitudes involved.

It's a 20 foot container, so it's 6.1 x 2.4 x 2.6 m (l,w,h). The largest aspect it can place to the sun is probably the aspect seen when looking down at about 45 degrees at the long side and roof, an area of (2.4^2 + 2.6^2)^-2 * 6.1 = 22 m^2. At peak insolation of 1kW / m^2 that's 22kW hitting it from the sun. Assuming an albedo of 0.7 for the white finish, 22 * 0.3 = 6.6 kW will be absorbed by the container.

Ignoring the ancillary systems and allowing a conservative 100W for each of the 252 1-U servers, we get a figure of around 25 kW of heat generated by the kit inside.

So it looks like solar effects aren't negligible, but at worst it looks like the sun would be contributing something like 20% of the heat load. Interestingly, Stanford's choice of white paint looks to be saving them over 10 kW of solar heating in the midday sun. Black paint in the sun isn't a very bright idea, Sun.

Outsourced?

[mcrbids]

I just don't get it. I mean, I really, REALLY don't get it?!?!?!

Places like 365 Main offer top-notch server hosting for dirt-cheap prices. I have a half-rack there with 6, quad-core Opteron clustered LAMP servers in place now. Reliability is excellent, bandwidth availability is fabulous (we have a Gb interface to the Internet) and the price is just astonishingly cheap - although we are an "Internet Company", we spend more on phone calls than we do on hosting and related fees. Never mind hotels and travel/flight expenses!

We've gotten between 4 and 5 nines of uptime over the last 4 years just by using quality software, (Linux software stack) good quality (but generic) hardware, and a good quality, outsourced hosting environment, at a price you simply couldn't believe.

I've seen plenty of other companies roll their own datacenter at 3-10x the cost, with greatly reduced reliability, embarassing outages, and lots of internal friction over maintenance overhead. Me? I want it cheap, fast, and reliable. It really IS a case of cheaper is actually the best!

I can see the need for an "internal" datacenter if your needs are largely local and your bandwidth usage is great - think enterprise application availability, or large volume local network storage. But if your company's core deliverable is public facing, I can't imagine any advantage to having your own datacenter until you company income passes the "100 million/year" mark.

And yes, I'm the CTO of a million-dollar per year Internet services/software company, growing at about 50% annually. (We'll probably be close to the 2 million-dollar-per-year mark by Christmas)

Or 672 blade servers and 5000 cores

[Paul Crowley]

You could fix six 6U blade units into each of the seven general-purpose racks, and put sixteen blades into each. Put two quad-core processors on each blade, get 5376 processors into the rack. That should put you somewhere interesting in the Top500.