When evaluating infrastructure-as-a-service offerings, the challenge is to compare the value proposition with what you already have in your own data center. The obvious wins for IaaS include so-called cloud bursting, where you need lots of computing power for a relatively short time. Obviously, building out infrastructure for peak loads is wasteful, so wise use of an infrastructure service for demand spikes can level out your costs. That's a no-brainer.
But that’s not the usual case, so understanding the cost benefit of IaaS is a tricky business. There have been many discussions lately of calculations that assess the incremental cost of using a system versus the cost of it just sitting idle in your data center. That incremental cost plus the baseline cost during the period of real use “feels” like the cost that should be incurred for a cloud service. Even if you paid a premium on top of that cost, you’d still be saving money, right? Well, maybe.
One problem is that calculating these scenarios isn't easy. So let’s take a very simple example. Say a company decides it will offer hammer as a service to carpenters. The conventional alternative, according to HaaS Inc., is that a carpenter buys a hammer, which is a capital expenditure--usually anywhere from $30 to $100--and then that expensive hammer sits in the toolbox, unused, most of the time. HaaS will provide a hammer just when the carpenter needs it, and it will charge just $0.001 per nail driven.
The first problem is apparent. How does the carpenter know if that’s a fair price? The answer is different depending on whether our tradesman is a framer who pounds relatively few nails, or a roofer who tosses down a piece of plywood and proceeds to hammer about 100 nails into it as fast as he can. In either case, the carpenter needs to measure current usage and estimate future usage over time. Likewise, without measurements of what they currently do and expect to do, IT pros have no idea if an infrastructure service is a good deal.
[ How does your cloud plan stack up financially? Check out our cloud computing ROI modeler. ]
Let’s say our carpenter can pound six nails per minute, but he typically pounds nails for only one hour per working day. At 250 days a year, that’s 90,000 nails driven per year. The HaaS bill would be $90 a year. Hammers make great Christmas presents, so even at one per year, the HaaS proposition is about a wash.
But the calculation would change radically if hammers were subject to Moore’s Law. In that case, every 18 to 24 months, they would cost half as much to produce and be able to pound twice as many nails in a given time. If carpenters can pound six nails per minute now, Moore’s Law says that in 10 years they’ll have a $2 hammer that will let them nail down an entire sheet of plywood in an instant. In 20 years, hammers would cost about a dime and could nail an entire roof in a flash. None of these performance improvements has anything to do with the service HaaS provides or on the skill of the carpenter using the Moore’s Law hammer. The hammer just does more and costs less.
My point is that the Moore’s Law advantage is immense and isn’t something you should give up lightly, but some cloud providers are asking you to do exactly that. Let’s hope our carpenter didn’t sign a five-year contract! Cloud providers take different approaches to the pricing problem. For its EC2 service, Amazon bases costs on what it calls the elastic compute unit. Back in 2006, it based the unit on a single Xeon core running at 1.7 GHz. From that base, prices go up depending on myriad factors, including memory needs and the term of the contract, three years being the longest period. Amazon updates its model over time based on newer processors.
There’s been some crowing about the fact that over the five years it’s been in this business, Amazon has cut prices more than 40 times. Well, I should hope so. Given the improvements in processor performance and storage capacity per drive, customers should expect--no, demand--that prices continually fall.
In fact, the U.S. Department of Energy, in a recent report comparing the cost of the government's in-house supercomputing centers to commercial cloud services, concludes that Amazon pricing isn't keeping pace with Moore's Law. "For example, the cost of a typical compute offering in Amazon has fallen only 18% in the last five years with little or no change in capability," the report maintains. "This translates into a compound annual improvement of roughly 4%. Meanwhile, the number of cores available in a typical server has increased by a factor of 6x to 12x over the same period with only modest increases in costs."
If you don’t hold your cloud vendors feet to the fire on pricing, you’re giving up the fundamental advantage that IT has had and exploited for almost four decades. That might be OK, but you’d better get something incredibly valuable in return. Make your infrastructure-as-a-service provider prove that value to you.