Slashdot Mirror
The Soaring Costs for New Data Center Projects
miller60 writes "The cost of building a quality data center is rising fast. Equinix will spend $165 million to convert a Chicago warehouse into a data center, while Microsoft is said to be shopping Texas sites for a massive server farm that could cost as much as $600 million. Just three years ago, data centers were dirt cheap due to a glut of facilities built by failed dot-coms and telcos like Exodus, AboveNet and WorldCom. Those sites have been bought up amid surging demand for data storage, so companies needing data center space must either build from scratch or convert existing industrial sites. Microsoft and Yahoo are each building centers in central Washington, where cheap hydro electric power from nearby dams helps them save on energy costs, which can be enormous for high-density server installations."
Some of these firms should really start looking at warehouses in Detroit. If you can secure the facility properly, you can get TONS of old warehouses and factory floors for very little. Look at the conversion that Wayne State did with techtown - they converted an old abandoned warehouse into usable high-tech space (and the real estate was virtually free).
- tom -
In the finest of Slashdot traditions I'm speaking from barely informed ignorance here:
It seems to me you can control your costs by buying existing space, like a mothballed factory, in an economically depressed area. Like, say, anywhere in the rust belt. You've got a bit of flexibility in siting as long as you can get Internet pipes, and you don't necessarily *have* to set up in an area known for a workforce with a high degree of tech skill (and absurd prevailing wages along with almost certainly having higher cost of everything because its metropolitan).
Our technology incubator in Japan is in a park with a few major data centers and is located 40 miles from the middle of nowhere. The US analog would be siting the datacenter in a cornfield in central Illinois. We have (comparitively) cheap power rates, a cost of living (and prevailing salaries) a fraction of that in Nagoya, and the rent (heavily subsidized by local government, which may not be an option for folks discussed in these articles) is a song.
Help poke pirates in the eyepatch, arr.
I work for a large financial institution.
We have a LOT of data...and not just account data.
Back in the 80's, the standard was two mainframes in the same room, back-up
tapes kept on and off site, and a contract with a company to supply a DR computer
if it was ever needed.
Cut to 2006...
We have dual fully redundant data centers, each with many mainframes, and pipes
big enough to drive a dump truck full of bits between the two.
A third one is about to open and a fourth is under construction.
Most of this is for SOX.
$7.95/mo, 200 GB disk, 2TBxfer, MySQL, PHP, RoR.
I grew up in Moses Lake, WA, which is about 30 miles SE of Quincy. This should be going into Moses Lake, but it isn't. We have goddamn fiber optics laid all over that town (and the county, exempting Quincy due to some sort of contract the PUD had with Verizon, I believe) going right up to people's houses. I enjoyed a 100Mbps symmetric connection for a while...then my bandwidth got capped. In fact, the PUD is charging the service providers so damn much for bandwidth, some have to cap it at 1Mbps down/512Kbps up. That's slower than fucking DSL and Cable! The local PUD is sitting on a fucking GOLDMINE and they're not doing a goddamned thing about it! They could have easily wooed MS and Yahoo into Moses Lake to build their datafarms there using the PUD's fiber network through the local providers (the PUD can't sell service, so they sell the use of the network to ISPs) and made things better in the town. But they're not doing shit. They haven't been doing much to promote, pitch, or package it for big guys to come in and build a major server farm out here. It pisses me off to no end to see those fuckers doing so little to help that town, and it needs all the help it can get. AAARRRGGGHHH!!!
The Westin Building (no not THAT office space) still has plenty of space, including the entire 5th floor!
The cost of turning that into a safe datacenter environment would be enormous. When was the last time you heard of a abandoned factory being built to hold a temperature controlled environment? The costs that go into making a real datacenter are significant, and building the place from scratch for that purpose can be cheaper. Building a datacenter right downtown is a stupid idea, but that doesn't make building it out in the boonies a good one.
"I use a Mac because I'm just better than you are."
Believe it or not, the speed of light is too slow. Latency would be an issue if the data center was in Alaska.
As we know worker moral is important, and considering the traditional living arrangements of your standard computer geek, it stands to reason that they should build their data center in the most awesome basement ever built! Hey, one can dream can't he? To the batcave!
Like I said: I live in Japan. We're the earthquake capital of the world, and yet somehow we manage to have buildings stay standing. Many of them also contain computers or millions of dollars of capital, strange as this may be. I trust that the folks living in Iowa and Detroit have figured out some combination of construction techniques, building codes, and insurance schemes which enables their cities to be something other than windswept wastelands. I mean, how long has the auto industry put billion-dollar factories in Detroit? And how many times have you seen GM say "Aww shootskie, we forgot about the ice storms and now three production lines are buried under 400 tons of collapsed roof and snow?"
Help poke pirates in the eyepatch, arr.
Actually your not that far off the mark. Building a data center basically comes down to five key components:
1. Getting lots of cheap power. Being next to a power plant with tons of extra capacity doesn't hurt. The farther you are, the more loss, and that means more $$$ per MW.
2. Internet pipes. Having X thousand servers up and running with nowhere to push the bits is pretty useless. I'm not sure if most people understand how hard it is to say get 40-60 Gig of bandwidth to the middle of nowhere. It takes months, if not years, to put in the right infrastructure. If you think I'm lying, call up say, Sprint and ask them for a 10GE pipe to the middle of Iowa but be careful, thier laughter might hurt your ears.
3. Cheap labor. Gotta have bodies to run everything from cooling to electric to security. You could do it with Robots, but after a while, as with all robots, they want to kill.
4. Favorable tax status. When you install hundreds of millions of dollars of hardware in one place, the amount of taxes you pay on it becomes non trivial. Many companies work out sweet tax deals with the local governments for just this reason. Basically the conversation goes, "You give us cheap tax treatment, we give you 500 jobs."
5. And of course, cheap land. Pretty self explanitory.
Because of small number of sites that are favorable in all these ways, I'm not surprised that costs have increased. But like all things this will change. Places with the last 3 bullets will build out power and connectivity and the datacenters will pop up all over the midwest. Either there or we'll outsource them to India.
ez
1. Access to a large body of water cuts costs immensely when dumping the heat from the beast. Fresh is preferred but not required.
2. Access to high voltage lines, or a short distance to one that can be tied into. 34.5 and 105kV lines are expensive to build and maintain on a long-term basis.
3. Access to fuel. Ideally rail, ship, or pipeline, because power plants burn massive quantities of fuel. Trucks do not cut it unless the distance is extremely short.
I recently worked at a power station that was originally built with none of these things. The only people to ever make any money from this white elefant were the contracters that built it.
Build your datacenter near a large body of water (or maybe in Juneau?). Build it near a power station (or build your own steam plant?). Build near some big strands of fiber. Being in the middle of nowhere for the sake of being in the middle of nowhere only profits the contractors.
Even those who arrange and design shrubberies are under considerable economic stress at this period in history.
It's incredibly uninformed to talk of costs in terms of total dollars!
The old metric was in $/sq. ft., and today it is better to talk in terms of $/kW given higher densities.
For a wide range of data centers, the building shell cost is around $100-250/sq. ft. An enterprise (EIA 692 "Tier 4") data center costs about $22k/kW, plus the high end of the building shell cost. A "Tier 3" data center is closer to $20k/kW and $200/sq. ft. When you drop to Tier 2, you cut the cost in about half, at $12k/kW.
The only costs that have risen dramatically recently are generators and copper, which have a one-year lead time for big engines typically used (1.5-2+ MW) for the generetor, and about triple the cost three years ago for copper-- maybe a 15% premium maximum for a large data center.
Costs get much more complicated when you talk about provisions for future expansion and site constraints.
As for energy costs, yes, cheaper electricity is good for a data center. A 2MW data center will save about $350k/year if they can drop their electricity cost by $0.01 per kWh!
They're naturally temperature-controlled: anything more than a few feet down is going to hover around 40-50F
Yeah, sure, until you saturate the heat capacity of the ground around you. Mind, the ground conducts slowly, and datacenters have a lot higher power density than your basement.
"We returned the General to El Salvador, or maybe Guatemala, it's difficult to tell from 10,000 feet"
Microsoft is willing to spend years and years waiting for the Xbox to become properly profitable, so it wouldn't be particularly surprising.
I love this cycle, where internet business heats up and companies start building datacenters to keep up with perceived demand. It happened the last time around, with companies like Exodus, Cable & Wireless, and all the others who were overbuilt when demand didn't materialize.
Anything over 50k sf of datacenter is more than enough, assuming you've got cheap and available power, and close to a couple fiber loops. The big reason that these new datacenters are so large (200k-400k sf, compared that to 1 floor of a high rise office at 30k sf) is because they aren't allowed to have the power density (elec co can only supply so much at reasonable price). With servers more power hungry, yet smaller, there's a need for more power/cooling, but less space.
Building new isn't all that different in cost of retrofitting an old warehouse. I'd just buy one of the small operators out there and be up and running for a % of the cost. The problem there is that there's a company called Digital Realty Trust buying all a lot of the datacenters in the market, and they've got a ton of cash.
So maybe the rust belt should be fighting for these developments, but they can't overcome 1 issue - companies want to be close to their datacenter. It goes against the security mission, the cost justification, and just about everything else; but these always get built right next to corporate HQ or in some metropolitan area. Doh!
Why cut IT when your office space costs $3/sf? gibso
The cost of building out datacentres has been soaring for several reasons, the first real issue is being able to provide enough power for todays power hungry servers to run at any sort of density required to actually churn the data. We see datacentres only able to offer very low amount of power per square meter in the UK, which is very low and can often only provide you with upto 4 quad processor XEON servers per rack. When you can only have that density the cost is much greater. The other aspect is how do you cool it, the traditional airconditioning raised floor method really does not work as its almost impossible to actually cool where you have to cool and there will always be hot spots even if your doing warm row cold row designs etc. Its important to seal the cool air in and funnel it to where its needed. APC have recently been working heavily in this area and claim to be able to cool MASSIVE amounts of density. The other issue is the management aspects of these datacentres. In the past you could design a datacentre to be good for 5-10 years now its hard to design something which will be good for the same length of time becaues the power requirements, cooling and power grid is something that is often over utilised (as per in the UK datacentre market).
This is the price of AJAX. If users are constantly going back to the server in the middle of a page, you need more server capacity. Really, the AJAX approach is a hideously inefficient way to update a form. We're now seeing the price of that.
Why not build your datacenter in alaska where it's colder year round.
Our datacenter in just up the Hudson from NYC and was built back in the 1960s, When IBM Ruled The Datacenter, and disk farms generated a lot of heat and the ambient temperature needed to be roughly 70F.
So, the DC is in the 2nd basement, and (had) vents to the outside, so cold winter air could be shunted into the room.
Became obsolete, though, in the mid 1990s when the huge 3390 farm was replaced by a couple of EMC cabinets and the bipolar mainframe was replaced with a CMOS s/390.
I'd have thought building the thing in Texas would just help pump up your A/C costs.
Depends on how well it's insulated. When the building is gutted, that's the perfect time to spray on insulation.
"I don't know, therefore Aliens" Wafflebox1
French Iliad SAS (the parent company of the ISP Free) just started a cute server renting business in a former Exodus datacenter.
The point of interest is that servers are fanless, built on low-consumption VIA processors, and consume about 20W/server.
That should make the cost of operation much lower than traditional hosting...
See details on http://www.dedibox.fr/index.php?rub=offre (in french)
Pictures of the datacenter: http://www.dedibox.fr/index.php?rub=datacenter
The cost of turning that into a safe datacenter environment would be enormous. When was the last time you heard of a abandoned factory being built to hold a temperature controlled environment?
oh for crying out loud. It amazes me the lack of thought outside the box people have.
Options...
1 - spend very little and build seperate enclosures inside the wearhouse that hold the libert units for environment control and the servers in data-center pods.
2 - go uber cheap. Buy a bunch of camper trailers that are gutted and put the servers inside those parked in the wearhouse. works great and I have seen several startups that did exactly that. this also works very well for rental property as you can pull up stakes and move your datacenter within minutes of getting your data pipes into another cheap wearhouse.
the best option and the one usually does in these types of datacenters is the first. you can hire simple general contractors to build interior walls with roofs that are only 10 feet high and insulate the crap out of them to make the perfect datacenter within 5 - 30 days.
It's the mentially retarted CEO's and Venture Capilolists that think you need to spend 80 million dollars on a flashy facility with lots of glass and artwork and special "touches" that only impress clients that will never go there or see it.
Do not look at laser with remaining good eye.
Clearly a case of Intelligent Design.
I've seen a number of conflicting estimates on how much power computers and digital devices use.
One source decries widescreen TVs as the "SUV" of the 21st century . The average plasma TV consumes more power per hour than the average refrigerator, the previous household energy hog.
This comment cuts across several threads on costs.
Costs alone are not enough. What is needed is a unit cost. For example, is unit cost per user rising or falling? If it is falling but the user base is growing rapidly, you are getting a good deal even though costs may be increasing.
Also, things such as redudent server, backups, power backups etc. should probably be counted as an insurance cost and measured against cost of down time. If the cost of downtime increases much faster than the cost of this 'insurance' then you are probably getting a good deal.
To say 'costs are rising' without a benefit analysis is meaningless.
Also, I wonder how much of this is due to bloated apps and poor design (XML anyone?). Is this explosion in servers due to crappy code and bad data models. I suspect some of it is though it has to be looked at on an application-by-application basis.
And while I am on the topic, multi-tier does *not* mean multi server. I have no idea how this myth got started (hardware vendors maybe?). You can, if you like, run all tiers on one server if your code is not leaky. For security reasons you probably should put your web server on its own box, but then if you have 5 tiers and a DB engine there is no reason why a good server can't run all of them in most cases. Unless, of course, the code is crap.
My semi-informed opinion....
putting the 'B' in LGBTQ+
Should be using VMware Infrastructure 3 :)
My company is building a new DC in Texas too. We are doing it on our existing campus by gutting and renovating an older building but the costs are still going to be huge.
In the meantime, I've been building one of the first VMware ESX environments our company has ever used. It started out as a simple 6 host server environment but has grown to over 20 DL 580s and 585s hosting hundreds of Virtual Machines. The initial investment is high but the operating costs are lower, the cabling costs are lower, the HVAC costs are lower, and of course, a VMware host server takes up less real estate.
If my company had focused on VMware, or virtualization in general, early on, they wouldn't need three datacenters and they wouldn't be building a fourth.
There is nothing inherently safe about liberty. That's why so many people died protecting it.