Data Center Building Boom In Silicon Valley

1sockchuck writes "Data center developers are building like mad in Silicon Valley, with seven active projects in Santa Clara alone. The building boom includes the resumption of several stalled projects that prompted concerns of a shortage of wholesale data center space in the Valley. The flurry of construction activity is different from the overbuilding during the dot-com boom, which was characterized by too much funding and too few customers. This time, industry experts say, the end of a funding drought has created a situation in which construction is struggling to stay ahead of demand from companies like Facebook — which just scarfed up an entire new data center in Santa Clara."

Comments

sure sure

[timmarhy]

"The flurry of construction activity is different than the overbuilding during the dot-com boom"

thats what they all say.

what about when the next fad comes along and facebook is forgotten over night?

Re:sure sure

[nacturation]

what about when the next fad comes along and facebook is forgotten over night?

If you subscribe to the theory that Facebook has built demand, then that demand (with the corresponding need for servers) will shift elsewhere. If it ends up being that a large part of the demand simply vanishes, then yeah... they will have overbuilt.

Re:sure sure

[William Robinson]

"The flurry of construction activity is different than the overbuilding during the dot-com boom"

thats what they all say.

what about when the next fad comes along and facebook is forgotten over night?

Maybe. But it could be true as well, if you kind of believe in Gartner's hype curve. I do, because, I have seen many things going through that phase of disillusion and pick up again when time comes. Maybe dot com is going to become part of steady growth.

my 2 cents.

Re:sure sure

[Surt]

The main reason to locate in CA is latency.

Re:sure sure

[Trepidity]

I hear that argument, but is CA really some sort of global optimum for latency of a data center? It's a reasonably well connected location, but I can't help but think it's being somewhat overvalued because many of the engineers are in CA, and it lowers latency to them. But most users are not in CA, and isn't latency to the end-users the main issue?

No real space shortage

[parallel_prankster]

Nobody wants to pay the ridiculous land prices for storing machines and blowing cold air on them. If you come to Silicon valley you will see that there is so much space everywhere. But real estate is crazy expensive. They could totally build data centers a little away from here, it may just be easier for them to have data centers closer for reliability/availability etc etc purposes.

Strange move

[sunderland56]

So, people are constructing new data centers on some of the most expensive real estate in the USA, in an area with highly paid IT workers with zero company loyalty, and an area of high electricity rates. Note to self: do not invest in these companies.

Re:Strange move

[timmarhy]

so who builds the systems, designs the floor layouts, maintains the hardware?

Re:Strange move

[evilviper]

So, people are constructing new data centers on some of the most expensive real estate in the USA, in an area with highly paid IT workers

The price of the land is pretty trivial, as relatively little is needed, and lots of money will be made with it over the next few years.

IT workers are highly paid as a side-effect of being the most highly skilled people available. Put a data center in Oklahoma, and you'll find some nice cheap IT workers, who have very little idea what they are doing. In a competitive market, the employees have to be just as competitive as the employers. The ample supply of highly-skilled labor is exactly why companies want to be there.

with zero company loyalty

See above. "Company loyalty" is actually a negative symptom. All those I've seen who have been employed at a company for a decade or more, do so because they are sufficiently incompetent to not find better pay or challenges elsewhere, but are just good enough to provide some value to the company.

There have been many papers written on the fact that, as pay increases over the years, the relative cost/benefit to employees goes down. Short-term employees is actually a preferred option. And frankly, if companies needed or sufficiently wanted employee loyalty, they just need to reverse the past 20+ years of taking away all benefits, but they'd rather not do that. Only a fool is loyal to a company in this day and age.

and an area of high electricity rates.

We're talking, what, 50% more expensive than the cheapest reliable electricity in the country? While it's not the cheapest, it isn't terrible.

Re:Strange move

[Anonymous Coward]

Note to self: Invest in the disaster recovery industry

Re:Strange move

[Trepidity]

There are more total tech employees in Silicon Valley, but I haven't noticed any higher average quality as compared to Boston, Altanta, Austin, Seattle, Portland, or other such places.

Re:customers

[oldhack]

"The problem was those customers were ultimately not viable. "

But Facebook is completely different.

Hehehe.

Re:I'm not impressed

[ricky-road-flats]

What are *you* on about? 3kW is a lot, when it's 24x7x365. Add it up. The house you mention is very unlikely to add up to anywhere near 3kW constantly for the whole year. The comparison (and your other about the desktop PC) is insane, and here's why:

His is one rack, in a row of dozens, with (unless it's a very small datacentre) dozens of rows.

All the customers in all the racks are trying to maximise their utilisation of the rack (extra racks cost more), and the utilisaton of the systems in those racks (more computers cost more). Each of those hundreds of racks needs multiple kW (some more than 3, not many less than 2 or so), with huge reliability.

Now add nearly the same amount of power again to cool all those racks, to keep air passing over them all. As well as powering the chillers, and driving the air down many channels to get it to every intake fan in every rack, all of which needs to have very good filtering (usually HEPA), and add on dehumidifying on top.

Multiple feeds into the building from the grid, UPS protection, surge protection, switching between live feed 1, live feed 2, and UPS power all has to be seamless enough to not bother a nodern computer - it all adds up to a very hard job. Yes it's an established process, but that doesn't make it easy.

Your midrange desktop with 4 cores - you couldn't get more than around 20 of those in a single rack, and you would be drawing way more than 3 kW to drive them. To get 200 cores and storage (and presumably some network kit too) into a single rack is still impressive now - and for it to draw only 3 kW is impressive - they must be very efficient units.

Add in the hard drives, RAM, fans, lossy power supplies, chipsets, switches, etc.

Oh, and very few professionals would use 2 TB SATA drives in a datacentre setting. Most units nowadays use 2.5" drives, and in the SAS world that limits you to 300 GB fast ones or 450 GB slow ones - 600 GB has been announced but it takes a while to become actually used. You need more though, as you need RAID to protect against failures. That frequently means installing double what you need in terms of raw storage. Then, you throw in a few hot spares for good measure. It all adds up.