When IaaS Won’t Cut It
When I was at CeBIT 2011, I remember hearing a speaker who talked about how easy it was in today’s world to start a business. He had a slide with a picture of someone using a laptop on an empty beach — the image being you could come up with an idea anywhere and get started.
With just a credit card, you could have immediate access to infrastructure services. It’s a golden age for entrepreneurs — except when that vision doesn’t work.
For many businesses having access to services like Amazon EC2 gives them an instant scalable data center that starts off inexpensive while they’re small and scales as the business grows. Ten years ago if an entrepreneur had a vision for a data-heavy business, it required substantial capital to get started to build out a data center, putting their idea out of reach for many.
Today, as the speaker at CeBIT pointed out it’s a simple matter of pulling out a credit card, but what about when your requirements are a bit steeper than what Amazon provides you. Oh, you can shop around for a higher end Infrastructure as a Service provider — I’m betting someone must offer boutique services — but as I wrote yesterday in my post, Zynga Shows How to Build a Private Cloud, Zynga‘s CTO Allan Leinwand sees Amazon EC2 as a fantastic service, but one that has its limitations because in the end it’s a generic set of services designed to meet the needs of many different types of users.
There’s nothing wrong with that. It works for tens of thousands of users, giving them access to servers, storage and RAM for a reasonable price. But there are situations where those generic services don’t cut it. That’s one of the reasons Zynga decided to build its own data center and use Amazon IaaS services in a lesser role.
It’s also why blekko CTO Greg Lindahl writing on the High Scalability blog explained why his company needed to build its own rather expensive data center to start the search engine because of its rather extraordinary hardware requirements. The scale is amazing requiring several thousand petabyte disks, but that’s just the begininng. Quoting the post:
“Serving query results quickly involves having most of the index in RAM or on solid state (flash) disk. If you can buy a server with 100 gigabytes of RAM for about $3,000, that’s 1,000 servers at a capital cost of $3 million, plus about $1 million per year of server co-location cost (power/cooling/space.) The SSD alternative requires fewer servers, but serves a lot fewer queries per second, because SSDs are much slower than RAM,” Lindahl wrote.
He goes onto say why Amazon doesn’t make it for him, starting with the the fact that Amazon lacks SSDs and putting the index in RAM gets costly for a small business. Further the way the Amazon charges for usage would make it less cost-effective for Blekko to go the IaaS route. In the end they bought 700 servers to start. Today they are up to 1500.
So while many companies can spare the capital requirements of building a data center by using Infrastructure as a Service, not everyone can — and it depends a great deal on your own business and its unique requirements.