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 worked for a very small company that had a server-rack in a cupboard without ventilation. In winter we'd open the door to keep the office warm. In summer we'd keep it closed to stop making the office to warm - there was no air conditioning. The temperature must have varied from 16 degrees C, maybe lower at night to 35 degrees C and the server never had any problems.
I also worked with someone who worked night shift as an operator in a large company that did have an air-conditioned computer room. During the day the machine room was treated with reverence, carefully dusted with special cloths, etc.. He told me that at night when they got bored they'd play cricket down the central corridor with a tennis ball and a hard back book. The computer cabinets regularly got hit with the ball and once or twice had people run into them. On one occasion a disk unit started giving "media error warnings" but apart from that no ill effects again.
Still, it's an interesting approach even if you're *just* dodging bullets and this is a disaster recovery scenario for your company. If anything it proves that you don't need a white-room, halon-protected, perfectly air conditioned data center to run your business, which seems to be the common belief across the US, European and Canadian enterprise.
Just ask any of the companies in the Gulf area affected by Ike if they would have been glad to have something like this in place a month ago.
I could have told them that computers tend to be resilient. I ran lots of them for many years in a little room at ambient temperature or higher, and high humidity. Every time I opened one of them up to upgrade or something I was amazed that they would even run at all. And the dirt...
Web2.0: I love when people Flickr my cuil and digg my boingboing until my google is reddit and I start to yahoo
You are talking about the other kind of datacenter.
Regarding this issue you have 2 kinds of datacenters:
- the cluster/cloud type where servers are expandable. They might die but you don't care because you have loads and all your data is redundant (e.g. Google, most nodes of a cluster, web servers etc)
- the big iron kind where you buy high quality machines, support, redundant power supplies, redundant NICs, pay people with pagers to babysit them, lower the temperature to increase the MTBF etc.
All this research applies to the first case. You are right to pinpoint that in the second case you will still want to take all the precautions you can to avoid failures.
Agreed. This is more an ad for the servers (HP) than it is for software (Windows). If these servers didn't have a hardware failure in many months outside, that says a lot about the servers. I know my server at home couldn't do it no matter what OS I used.
Well, there's spam egg sausage and spam, that's not got much spam in it.
I'm pretty sure my SAN's redundancy has nothing to do with servers attached to it dying.
With the July heat, it's not just the baked electronics in the servers, either. Your hard drives become less and less reliable, and their expected lifetime is far shorter after they've operated for any length of time in conditions like you're experiencing.
You also completely ignore the cost of the downtime itself. Doesn't matter how much it costs to restore the data if you're down long enough that your clients lose faith in you and leave.
"Good will" is on an account sheet for a reason.
Check out my sysadmin blog!
> suggesting that servers may be sturdier than believed
Anyone who thought you couldn't run a reliable server in 85+ degree heat in a sweaty, humid room with water dripping on a *sealed chassis* was a moron anyway. Most servers come with filters on the fan vents, are pretty tightly sealed shut
otherwise (and none of them would vent out of the top of a chassis because it would impact the servers above and below, so where's the water going to drip into?)
Air conditioning and all the other niceness we get in server rooms is just an insurance policy.
Good point, these are not Pacific Rim parts-in-a-box clones. These are DL585s. Hot-swapping redundant power supplies, redundant ROM, ECC memory and redundant NICs. These are the boxes that you're supposed to only power-up once, they should run forever.
If you're going to run them hot, it's best to have them outside because when all those fans go high, it's enough to wake the dead.
"The ferrets, they're every where I tell you!"
In my experience, when cooling systems fail in datacenters, servers don't fair that badly. If there is a weak hard drive or power supply that was close to failure before the cooling issues, they would most likely fail, but for the most part the servers could endure rack temps over 100F.
What usually took the biggest beating and had the higherst failure rates were networking equipment(hubs, switches, routers, fiber switches and repeaters, etc). Fiber equipment being the worst of a bad lot.
WARNING!! Opinion follows-->
I always thought that most datacenters were over designed. I think a lot of the specs are hold overs from the IBM MF days when the big iron used to put out more BTUs than a small stove and was susceptible to even the smallest contaminants and moisture. The latter was due to the more mechanical nature of the older tape drives.
I've had my server outside on the balcony for a couple of years here in Norway. No problems at all - temperatures ranging from +40C to -20C:
If you're buying DL585's then it's likely that the applications hosted on it have a downtime costs per hour (if not minute) equal to the cost of the hardware. These are the servers that large sections of the NYSE run on for instance.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
downtime costs per hour (if not minute) equal to the cost of the hardware
No argument there, and I'm not stepping past the potential losses in terms of business impact, I'm simply pointing out that the cost of the machines themselves aren't exactly negligible (regardless of how those costs stack up compared to other costs). As you probably know, there are people within an organization who get their butts chewed when a service stops running on a box and there are also those who get their butts chewed when a piece of hardware fails. I'm saying that just because the potential loss in revenue during the outage may be higher than the cost of the box, doesn't make the thousands of dollars invested in that box immaterial, cause someone's going to have to answer for it.
Just my 2 cents.
If you had actually followed the link back to the original blog post - unusual for slashdot, I know - you would have found out that the Microsoft employee that came up with this actually is a hardware engineer. If you had decided to indulge your curiosity a little further by googling Christian Belady, you would have found out that he had worked for several years designing servers at HP, with an emphasis on power efficiency. According to this article, he has several patents to his name. So yes, I would say he is qualified to tell hardware manufacturers what their systems are capable of. But don't let the facts stop you from getting in another jab at MS.