According to the economical side of Moore’s Law, processing power gets cheaper every year due to vendors being able to pack more of it in the same amount of space. Should cloud computing follow Moore’s Law?
Let’s take a look at Amazon’s computing offering in the cloud — the Elastic Compute Cloud (EC2). EC2 is a web service that lets customers rent Amazon servers on which they can host their own applications. There are different price levels called “instances.” The basic one is 10 cents an hour and, according to Amazon, is equivalent to a 32-bit system with 1.7 GB of memory, 1 EC2 Compute Unit, and 160 GB of instance storage.
So what is an EC2 Compute Unit? According to Amazon, it is equal to a 1.0-1.2 GHz 2007 Opteron or Xeon processor. When EC2 first came out in 2006, one EC2 Compute Unit was equivalent to an “early-2006 1.7 GHz Xeon processor” according to Amazon documentation on EC2 Instance Types.
(Two odd things about this: There’s a 20% difference between a 1.0 GHz processor and a 1.2 GHz processor, so what gives? And if a 1.0-1.2 GHz processor in 2007 is equivalent to a 1.7 GHz processor in 2006, why change the definition at all?)
The price has stayed the same since 2006 at 10 cents an hour for that basic instance, but you are getting the same amount of processing power now that you were in 2008. So it is not following the economic portion of Moore’s Law.
“You can now get a quad-core server for the same price you could get a single-core server in 2006. But cloud computing is not taking advantage of Moore’s Law,” said Raj Dutt, the CEO of Voxel Dot Net, a New York-based hosting company. “It’s the same price for the same amount of processing power.”
The question is whether it should. Clay Ryder, president of analyst firm Sageza Group, doesn’t necessarily think so. He sees EC2 and other cloud computing products to be based on a different pricing scheme. Whereas servers are based on a sales model that includes the cost of time, materials and markup, cloud computing is more of a values-based pricing model, and the two are not the same.
Ryder likened it to owning a car compared to renting or leasing. When you buy a car, you’re paying for the cost of materials to build it, the cost of labor it took to build (time), and any markup to make profit. When you rent one, you pay for a service.
“There is a lot of value in the Amazon approach,” he said. “You can turn it off and turn it on, and there’s no long-term costto you, and that is an intangible value.”
Ryder hits on something here. When you pay for EC2, you’re not just paying for the server hardware. You’re also paying for the data center infrastructure around it (land acquisition, building costs, power generation, chillers, racks, etc.) and the cost of labor it takes to maintain that infrastructure and the servers. If you have your own data center and your own people, you don’t pay for that when you go buy a dozen servers from Dell.
Some might still argue that at least a portion of the cost should be subject to Moore’s Law. After all, Amazon is charging the same price for the same amount of processing power, even though that processing power is getting cheaper for them to buy.
Then again, maybe Amazon has already factored in Moore’s Law, but has also factored in the increasing cost of labor, electricity, and materials to build a data center to run those servers. So in the end, it’s all a wash, and the price stays the same.