33 Comments

Summary:

The cost effectiveness and convenience of outsourcing your data center to a hosting provider has been under attack from a lack of consumable power from utility companies. There are numerous stories about major hosting providers turning away customers because of their power requirements. Some of the […]

The cost effectiveness and convenience of outsourcing your data center to a hosting provider has been under attack from a lack of consumable power from utility companies. There are numerous stories about major hosting providers turning away customers because of their power requirements. Some of the largest enterprises and service providers are have resorted to building their own data centers in remote locales where utility companies can provide the necessary power.

Outsourcing to a remote location may solve the power consumption problem, but can add travel expense and make maintenance difficult if every visit to the server rack requires a screening from the TSA. The question today is: Could solar power step in and save the day?

When you outsource your data center, you generally pay for space by usable rack-unit. Yet, because of the power consumption limitations of the hosting facilities you often can’t use all of the rack-units you have rented. For example, you may rent a server rack that has 50 rack-units that could hold 50 one rack-unit servers, but these servers will often consume more power than is allocated to each rack. To solve the issue, you end up renting more racks and distributing the servers over multiple racks. Blade servers and servers that consume less power can help, but there are plenty of power-hungry pieces of equipment still in production. So, in many cases if you want to outsource to a hosting provider that is easy to get to, you are stuck with paying for more racks than you can use.

To resolve this issue, I have been speaking with entrepreneurs that are working to solve the problem using solar energy. The theory is that if you can generate enough solar energy you can sufficiently supplement the utility company power and then locate a hosting provider in locations that are desirable to enterprises, such as downtown Manhattan or San Francisco.

Dissecting this theory a bit, modern hosting providers today are being built to serve 300 watts per square foot. Assuming a 100,000 square-foot hosting facility, that is a massive power consumption of 30 megawatts. If you build a 100,000 square-foot solar energy system on the roof of the hosting facility, my friends in the solar energy business tell me that would generate a total of only 1 megawatt. Yet, a solar energy system is geared toward shaving power during peak hours only, so you may only need 10 megawatts. That is still far more power than possible using solar energy.

But, combine solar energy with some of the latest battery storage technology and you should be able to build a 10 megawatt power system – you use the 1 megawatt solar system in combination with off-peak lower-priced power to charge the 10 megawatts of battery storage. There are lots of logistical issues here, such as the actual usable area on a roof and so on, but this seems realistic in theory. I’ve been told that such a system could cost up to $10 million to build and deploy (before the government rebates and tax credits).

The approach sounds quite novel and interesting – the real question to me is if the additional expense can really alleviate the power consumption limitations in a geographically desirable location.

Allan Leinwand is a venture partner with Panorama Capital and founder of Vyatta. He was also the CTO of Digital Island.

You’re subscribed! If you like, you can update your settings

  1. Disclaimer: I am the owner of a hosting company with ~500 servers online.

    It’s not just power that is the concern. A concern that is on the same level is cooling. For every watt of power your server draws, you have to cool that watt, too, which effectively doubles the power you need, and halves your available space. The best way to solve this is with more efficient servers that don’t generate as much heat. You’ll get a much better ROI right now by upgrading to servers that use much less power on average than you will by installing a fancy solar-power system on the roof.

  2. Blade Watch » GigaOM » Time for Solar Power Data Centers? Sunday, March 4, 2007

    [...] GigaOM » Time for Solar Power Data Centers? The cost effectiveness and convenience of outsourcing your data center to a hosting provider has been under attack from a lack of consumable power from utility companies. There are numerous stories about major hosting providers turning away customers because of their power requirements. Some of the largest enterprises and service providers are have resorted to building their own data centers in remote locales where utility companies can provide the necessary power. [...]

  3. Allan Leinwand Sunday, March 4, 2007

    SlashChick – You’re absolutely right that cooling is another major concern (and it takes power to run those A/C units). For this piece, I chose to focus strictly on the power issue and think that I took the power for the A/C units into my 300 watts/sq ft budget.

    I have seen some innovative data center design using hot/cold rows and efficient exhaust and recirculation units that can help reduce power requirements of cooling systems. Also, some of the more progressive blade computing systems do a good job of moving heat away from the server components.

    Thanks for the comments – ready to use solar in your data center?

  4. Isabel Wang Sunday, March 4, 2007

    Hey Allan,

    Is it absolutely necessary for servers to occupy prime NY/SF real estate? In addition to higher power prices and more limited capacity, the cost of building/operating facilities is an issue too.

    In the article below, a data center site selection consultant said he sees jobs leaving major metro areas for Sioux Falls or an Ames, Iowa, where real estate and personnel costs are much lower. That’s another approach for cutting the TSA out of one’s server management equation.

    http://searchdatacenter.techtarget.com/originalContent/0,289142,sid80_gci1238054,00.html

    At the moment, given the $10M price tag you mentioned, I’m guessing solar power might be more of a green choice (by http://www.aiso.net, for instance) than a cost-competitive option.

  5. Tim Aaronson Sunday, March 4, 2007

    Does every unit have its own power supply? Or every rack?
    Maybe there is, but if not why not centralize the DC power source and just run DC throughout the building? Would be better than every unit inefficiently generating its own DC from the AC.

  6. Solar power is about 1.4kw/m^2 … with aiming and conversion inefficiencies you are down to 10-20 watts/sq ft. Of course you need to deal with darkness:-)

    Solar is still far too expensive for this sort of application unless you have amazing subsidies.

    You need to find really cheap power … central Washington state hydro power is about as cheap as it comes. (no wonder Google and Yahoo have located there))

  7. steve gelmis Sunday, March 4, 2007

    This is an increasingly important topic, so thanks for raising it. And solar is a great idea, but as you’ve already established, solar is insufficient due to it’s relative lack of density. That should not become the end of the discussion though.

    My Manhattan-based ISP and telecom company runs a small data center, and we have researched this issue in depth. What we ended up with is a co-generation system.

    Instead of buying a diesel or natural gas generator for power back-up against grid outages — which becomes an unproductive sunk cost as soon as paid for — we purchased a micro-turbine co-generation system, and now make our own primary power using natural gas at a net savings per KWH to grid power. The grid becomes our back-up.

    The project cost came out to about the same as the generator option, but we’ve significantly lowered both our operating costs (even net of maint contract costs) and our environmental footprint.

    We needed only 60kw, and are initially using only about 1/2 of that. In a MW-scale facility a natural gas fuel cell would probably become the technology of choice, and would accomplish much the same thing.

    The key to why co-generation is a big improvement environmentally (and the source of even greater operating cost savings than I’ve yet indicated) is that the heat exhaust from the turbine-electric generation stage is harnessed to drive our air conditioning compressors without consuming additional fuel. In effect, we’ve reduced our electrical load requirements by about 1/3 by eliminating air conditioning as an electrically-driven application (other than the circulation fans and controls).

    Had we owned the building we are in, we would have scaled the project up enough to serve the electric, HVAC and hot water needs of the other tenants as well. There are already a few commercial and condo buildings in NYC being operated this way, but word of the availability of this technology at such small scale is not yet widely known about by decision makers. It has a great future though.

    Solar and wind are sexy because they represent the pure forms of their type — totally fuel-free and environmentally beneign power sources. Co-generation is more akin to trading an SUV for a Prius.

    On the other hand, it was totally cost-effective vs. a reliability-only investment, works in urban centers (even as just a commercial tenant), and makes a meaningful difference today.

    Down the road a little, we may add a limited amount of rooftop solar to cover the now tiny amount of differential in our daytime usage when staff are present (our corporate offices are at the same location). Now that our 24/7 base-load is covered efficiently our remaining peak demand has become something solar actually can be expected to address.

  8. SocialStartups.com » Solar power supplements Sunday, March 4, 2007

    [...] Solar powered data centers [...]

  9. There are two separate problems you’re trying to solve. One is the issue of not having enough power. You can solve that with batteries and/or solar power. The second is the issue of the cost of power during peak hours. You can solve that problem with batteries.

    However, both problems are already faced by the power companies and by lots of their customers. If the batteries are cost-effective, why wouldn’t the power company, or any third party, implement it? It has nothing to do with data centers.

    In fact data centers are probably a poor place to benefit from them, since servers tend to be on all the time. You gain the most in businesses that use power mainly during peak times.

  10. Guys, you are missing a large part of this story. Yes, heat is a model, but that is not what this is about.

    If you went with the SunEdison model you could basically do this without any capital outlays. And save money at the same time.

Comments have been disabled for this post