Slashdot Mirror
Microsoft Innovates Tent Data Centers
1sockchuck writes "The outside-the-box thinking in data center design continues. Microsoft has tested running a rack of servers in a tent outside one of its data centers. In seven months of testing, a small group of servers ran for seven months without failures, even when water dripped on the rack. The experiment builds on Intel's recent research on air-side economizers in suggesting that servers may be sturdier than believed, leaving more room to save energy by optimizing cooling set points and other key environmental settings in the server room."
I'm sure it'll work fine untill someone strolls past, lifts up the canvas and walks off with the entire rack. Or accidently flicks a cigarrette but at the tent. Or....
But what you're really doing in a situation like this is dodging bullets, rather than proving that we overbuild environmental in our server rooms. We KNOW that excess heat, water, humidity, etc can kill servers. These are facts that cannot be ignored.
I understand the idea here but still, do you really want to tell your bosses that the server room got to 115 F in July and killed the SAN because you skimped on the air units?
PHB: Well we just put up all of our servers outside. And it looks great! Say, what is that truck doing? Why is it driving so fast through all the security points... omg !
Guns are for wimps... Use a crossbow.. this way you can pin them to their chair when you go postal.
Funny how the military and the Live concert people have been doing this for years, but microsoft innovated putting servers in a tent.
Do not look at laser with remaining good eye.
While it's certainly an interesting experiment there is no way I would run my companies servers in a tent, especially a leaky tent.
Anyone who has built a few machines knows that hardware can prove to be a lot tougher than many people think it is. We once had a server running for over two years that had been dropped down a flight of steel stairs a few hours before delivery (we got the server free because it was really badly dented and no one thought it would actually run).
There is a difference between the above scenario though and the one where a whole rack of servers is sitting in a tent. One decent tear in the tent could easily flood the tent. Tough as they are I can't see any server running with water pouring into it and this scenario would result in the whole tent going down in one go. If you have to have a hot spare for this situation it's probably just easier to put it in a real building or a shipping container.
I used to have a better sig but it broke.
And from the other side, I'm constantly telling people to clean the crud out of their machines. Just last week a co-worker brought in her boyfriend's machine because it "would not work". Two minutes of blowing out the dust in the slots (RAM, AGP and PCI) and it booted up just fine.
I'm in agreement with the "dodging a bullet" comment.
Just because it is possible that there might not be problems (unless X, Y or Z happens) is not the same as taking pro-active steps to reduce the potential problems.
Sure, their server handled the water dripping on it.
But then, you would NOT be reading the story (because it would not be published) if the water had shorted out that server. It would have been a case of "Duh! They put the servers in a tent in the rain. What did they expect."
With stories like these, you will NEVER read of the failures. The failures are common sense. You will only read of the times when it seems to have worked. And only then because it seems to contradict "common sense".
While I agree with the notion of bulletproof data centers, I think one of the points of all these experiments, is that if you can save $100,000 a month on A/C and environmental costs, at the expense of reducing the life of $500,000 worth of hardware by 20%, you actually save money, because you spend so much more maintaining the environment than you do on the hardware itself -- as long as you plan for hardware failure and have appropriate backups (which you should anyway). On the other hand, if your hardware is worth a lot more, relative to your expenses, or if your hardware failure rate would increase sufficiently, then this approach wouldn't make any sense. It's all cost-benefit analysis.
-brian
Yeah, what do the hardware engineers know who designed and tested the servers?
/sarcasm
They know what the servers will survive, not what it could survive.
When designing a machine to work from 10C to 50C and from 20% to 70% humidity, they don't deliberately design it to fail just outside that range. They just make damned sure it won't fail within those ranges (at least, not because of temperature or humidity).
Microsoft's software engineers can show them what the servers are really capable of, without even testing them out for all four seasons.
Sarcasm ignored, yes, Microsoft (or any of us willing to sacrifice a server for the cause) can indeed demonstrate that a server can live in a more harsh environment than intended. Because, as mentioned above, the hardware engineers didn't design the systems to fail just outside their spec'd range.
We (as a whole) tend to baby servers because they cost a lot... But the cost of maintaining a perfect environment for them far outweighs the price for the actual hardware; If you can chop that expense out of the budget for the 99% of your servers that don't strictly require five-9s uptime, the savings in TCO could potentially far outweigh the increased cost of more frequent hardware replacement.
This harps back to mainframe days. In order to keep your warranty valid, you had to strictly control the environment - including having strip recorders to PROVE that you hadn't exceeded temp or humidity limits. The reason was that the heat output of these ECL beasts was so high that they were teetering on the brink of temperature-induced race conditions, physically burning their PCBs and causing thermal-expansion induced stresses in the mechanical components.
Nowadays we are nowhere near as close to max. rated tolerances and therefore can open windows in datacentres when the air-con fails. However, the old traditions die hard and what was true even 10 years ago (the last time I specc'd an ECL mainframe) is no longer valid.
I'd suggest that if it wasn't for security reasons, plus the noise they make and the dust they suck up, most IT equipment could be run in a normal office.
politicians are like babies' nappies: they should both be changed regularly and for the same reasons
When I said there would never be any Microsoft servers running in my department, I don't think they quite got my meaning.
They weren't microsoft servers - they were HP servers. For some reason MS are getting all the publicity and credit from this article, although they've actually done very little to deserve it. The hardware that should get the plaudits - typical!
politicians are like babies' nappies: they should both be changed regularly and for the same reasons
Intel's experiment was useful, but not surprising to the vast majority of IT people who frequently put racks in closets, their offices, etc. because that's the only place we have for them. Intel is just giving us ammo for tell the datacenter guys that we don't really need them if they're charging too much.
Microsoft's experiment is simply ludicrous. For many obvious reasons, theft being the most obvious, no one would ever actually run a server on a tent unless you're on some scientific expedition to some place where there are no buildings and you're not staying long enough to build one and there's no bandwidth available, even by satellite.
While Intel is addressing the problem of physical space costing far more than the computers we can store in it, Microsoft is almost making light of it. Stupid Redmond bastards.