Colocation


14
May 09

If A Tweet Killed a Tuna – Energy Cost Transparency in IT

One of the keys to improving anything is having enough information. This has been widely discussed in environmental circles, and recent innovations such as the Kill-A-Watt and the awesome hack the Tweet-A-Watt have lead to a more widespread appreciation for just *knowing* the amount of energy your appliances, computers, and home entertainment systems are consuming.  In addition to being surprising, the reality is that all too often assumptions are made about where to focus effort to fix a particular problem – or worse, you don’t even know a problem exists. But what to do with this information? At home it’s as easy as putting your devices on a power bar – such as your home theater – and turning it off when you’re not using it. Having the data enables you to make a decision – the decision to save money because all of the sudden it’s tangible.

These kinds of details can be applied at a really big and small scales too. What if you could measure the amount of power went into making your car? The amount of energy each Google search takes? The amount of energy for every tweet? Would knowing a tweet kills a tuna make you think twice? Would it enable you to make better decisions about the products you consume? Would it allow your customers to make better decisions about their energy efficiency?

This can apply to the hosting world too. Computers currently use more energy than the entire airline industry, and that’s expected to double within the next 5 years. Data centers consume a whopping 2-3% of the power in the United States alone. Hosting companies charge flat rates for collocation, virtual servers, shared hosting, etc. Bundled into that are the charges for electricity, and the electricity required to power the cooling. Unless you’re really close to the physical infrastructure, there’s no way to measure how efficient the servers are, or how much power your server is consuming. If we could measure the amount of power a server uses then you could incorporate that into the pricing of the server, and display the information separately. As a hosting company you would be able to make better decisions about which hardware, software, etc to use. As a hosting customer, you would be able to choose locations that are more power efficient. A slew of other possibilities exist. Due to power deregulation and trading markets in many locations, what costs a dollar during the day might cost 10 cents in the middle of the night.

hourly-demand-in-ontario

Data centers are built for peak capacity, but there should be an incentive for customers to adopt more energy efficient solutions. Being able to measure (in)efficiencies also means that making decisions about moving to a container might be easier to justify.


8
Oct 08

Free Servers are Expensive

There are a lot of ways for companies to bootstrap, and even more companies and partners willing to lend a helping hand if you know who to ask, and where to look. One of the companies that’s willing to help is Sun Microsystems.

Through their Startup Essentials program, companies can get access to Sun gear and resources at heavily discounted rates. That’s great, because Sun gear is pretty much the best out there, and being competitive with other companies like Dell and HP on price will help get their technology into more Data Centers and up and coming businesses. But like all shiny objects, servers lose their luster. Even if they’re free.

In order to run a server you have to secure colocation space, pay for bandwidth, buy some switching gear and a router, and depending on your setup get UPS power. Sure you can get cheap switches to connect your high-end servers, but depending on your work load it would be like running a jet engine on bacon fat.

If you’re running more than a few servers you’ll probably need to get more than the standard ~3000W/Rack – especially if you have a SAN device. That means one full rack with extra power. Most older data centers don’t have the cooling capacity to handle todays dense and powerful systems, so your full rack will probably be half full (if you’re lucky), then you’ll have to get another rack if you want to expand. That involves more waiting. Average amount of time for a colo to provision a new rack? 4-6 weeks. You’ll also need to pay for installation, and every colo provides space on contract so you’re committed for a year. Time is money, and waiting 4-6 weeks to be able to expand your environment means you have less flexibility.

There are certain scenarios where getting colo still makes sense. If you’re running your own hosting environment *as a business* then obviously having the control over your environment is necessary. Companies with certain regulatory or security requirements will also need to stick with colo, but otherwise, why lock yourself into contracts and inflexible environments and hire extra staff to manage that operation, when you can just rent some servers en-masse and get super cheap bandwidth?

If you run the numbers you can get just as much if not more *power* for the same amount of money from companies like Joyent, Amazon, ServerBeach or Rackspace, and not have to worry about contracts. While it’s nice to have an awesome piece of kit, sometimes it isn’t worth the time or money.