It is deceptively difficult to calculate the true costs of cloud versus on-premises deployments, as demonstrated by a new total cost of ownership tool made available by Amazon Web Services this month.
IT pros say they will use the free Amazon Web Services (AWS) Total Cost of Ownership (TCO) Calculator to get a baseline idea of on-premises versus public cloud costs, but closer inspection of fine-grained details and assumptions made by the tool revealed some areas where the true cost of cloud remains elusive.
"I would use it as a reality check to make sure I'm asking the right questions," said Scott McKay, CTO of ZappRx, a prescriptions management software startup based in Boston that has struggled to decide between an on-premises infrastructure and going all in with AWS.
"Then I'd double check things," McKay said.
He is right to take a cautious approach, according to AWS channel partners and analysts who have reviewed the AWS TCO calculator.
No 'one size fits all' approach
For example, very small deployments are not a good fit for the calculator, which calculates rack and network switch costs as well as the price of server hardware to decipher comparisons that show on-premises infrastructure to be vastly more expensive than AWS.
Using the basic form of the AWS TCO calculator, then, a three virtual machine (VM) host with four cores, 32 GB of RAM and 1 TB SAN storage is priced at a total of $112,898 for server hardware plus overhead -- not counting other costs like network bandwidth and storage capacity. That figure, ultimately, is compared against a total cost of $11,144 for AWS based on three-year Reserved Instance prices.
I have yet to come up with a scenario where I can show internal IT costs to be better than AWS.
"For the midmarket, [AWS] probably did a pretty good job," said Jared Reimer, co-founder of Cascadeo Corp., an IT consulting firm located in Mercer Island, Washington. "Where the model clearly doesn't work -- and honestly should have errored out instead of giving a bad result -- is in the case of a very small business."
For a business with 10 servers and one cabinet, the AWS TCO calculator estimates a savings of 80% to 90% by moving to AWS, but such a small business in reality would probably reject that math, Reimer said.
"They would say, 'Look, we can save colo and we can save hardware costs, but we've got to have an IT guy -- that's inevitable -- so the cost savings aren't going to be 90%; maybe they're going to be 30% or 40%,'" Reimer said.
However, as deployments scale, those rack and switch costs don't repeat for every VM, which leads to a more accurate comparison for midsized companies or projects.
There is also an advanced form of the AWS TCO Calculator that takes into account things such as storage I/O and network peak to average ratios, but it, too, fails to account for some details that an existing Amazon Simple Monthly Calculator includes, such as Elastic IP addresses and load balancing.
Moreover, try the following as a scenario with the advanced TCO calculator: a 100-VM environment, where each VM that has two cores and 8 GB of RAM and a VM usage of 60%. (The host is called "Host 2" with four cores and 256 GB of RAM). It is attached to a 50 TB SAN with a rate of backup data growth per month set to 5%, and 100 Mbps of network bandwidth with a peak to average ratio of three.
This will yield network costs of approximately $56,000 on-premises, versus network costs at Amazon of about $4000.
"The numbers [for AWS] should be the $56,000 plus the $4000 -- my clients almost never turn off their Internet connection when they use Amazon," said Kyle Hilgendorf, analyst with Gartner Inc. based in Stamford, Connecticut. "In fact, the Internet connection often becomes even more important, and clients invest more there."
A good first step on a long journey
Ultimately, this is a useful tool for IT practitioners who know how to interpret and adjust the raw results as necessary, according to both Reimer and Hilgendorf, but neither one would recommend it be taken as gospel.
"[AWS] put a good foot forward, but there are things that they have to improve, and it does make their services look quite advantageous," Hilgendorf said. "I have yet to come up with a scenario where I can show internal IT costs to be better than AWS."
These scenarios do exist in the real world, however, according to Reimer. One client that had a 1000-node Hadoop cluster in AWS, for example, found a very cheap colocation facility, very cheap power, used "disposable" white box servers, and was able to build the same environment with better performance for a lower cost.
On the other hand, a startup with an influx of venture capital might be able to run its own internal infrastructure at a lower total cost than AWS, but that doesn't take into account the "intangibles" of cloud, Reimer said, such as allowing high-priced developers to focus on value-generating activities rather than babysit an on-premises infrastructure.
In the end, this kind of TCO calculator can be picked apart because cloud versus on-premises cost analysis is very difficult to do, Reimer added.
"The numbers just aren't as rigid and hard as we want them to be," he said.