Last year industry lobbyist Uptime Institute commissioned a McKinnsey report on the cost of Cloud Computing designed to inject FUD into the cloud debate.
The report from McKinnsey claimed that Cloud Computing is MORE expensive than the traditional, physical data center.
While McKinnsey gave this report at the Uptime Institute, who’s membership is dominated by data center operators and Managed Service Providers, cloud computing’s main competition, the report does make good points about the hype and the mistaken use cases touted by some cloud proponents. This is reminiscent of the level of hype/promises in the mid 1990 about the World Wide Web.
Many of the claims and counter claims being made about Cloud Computing were meaningless at that point as there weren’t enough practitioners to make debate relevant. Fast forward a year and Amazon’s AWS service is reported to be doing half a billion dollars in revenue per year. Obviously Cloud is catching on.
So we decided to see if that jump in Cloud usage was driven by Cloud becoming cheaper than co-location. Prices have adjusted a bit and concerns over security are getting solved or proven to not be problems after all.
In order to help shed some light on the value and risks associated with Cloud we will attempt to do a simple, apples to apples financial comparison between two popular alternatives for building a Greenfield data center using both “The Cloud” in the form of Amazon Web Services and building a physical data center.
Can a datacenter be built in the Cloud with economics similar to co-location?
Companies are being bombarded with hype about Cloud Computing. Large, incumbent vendors are re-branding their legacy technology as Cloud, while the media labels every technology under the sun “Cloud” in order to drive readership. It’s no wonder that the picture around cloud is murky at best. This not only creates confusion, it forestalls companies from making rational, timely decisions about if and when to use cloud computing. In order to help shed some light on the value and risks associated with Cloud we will attempt to do a simple, apples to apples financial comparison between two popular alternatives for building a Greenfield data center using both “The Cloud” in the form of Amazon Web Services and building a physical data center.
Scope
This analysis examines the potential for using Amazon Web Services (AWS) as a virtual data center with building a traditional physical data center in a colocation facility such as Savvis, ATT or IBM. To keep things simple we have not included a third option of locating and building a data center at company owned property as that alternative adds the costs of real estate, power, cooling, taxes and physical security to the equation greatly increasing costs over either of the first two approaches (but that is certainly possible to do and has the benefits of ownership and physical control.)
We have also not included storage, data transfer, CDN, monitoring, geo-location and other extra features as direct cost comparisons are difficult due to the difference of buy vs. build for most of them (while one can buy these features from AWS or build them yourself, it is difficult to calculate the cost for the latter and, therefore, to make a fair comparison between the two.)
Assumptions
We’ve used as an example a medium sized data center model comprised of 1000 servers. The assumptions are that all the servers live in one datacenter located in a single US geography with no failover or disaster recovery. I’ve assumed there are 40 servers per rack for a total of 25 racks and that the colo facility will charge $1000 per powered rack per month. The average cost to buy a quad core commodity server is $2500. I’ve also assumed that networking gear will equal 20% of the server budget. We also assume that procurement for the colo approach will be spread equally across each quarter of a fiscal year in order to better control budgets and to give operations time to ingest the hardware and that all of this capital budget is absorbed in the first fiscal year. We have also assumed that, as a result of an AWS feature called Elastic Load Balancing – the ability to automatically bring servers up and down based on load – the absolute number of servers needed in the AWS approach is half of the colo approach. This may be conservative as physical servers typically run at 25-30% of their capacity. It is also common to provision physical data centers for “worst case” scenarios that can only be estimated. Lastly we’ve assumed that, in a cloud based datacenter, admins can better leverage automation tools and so have a two to one advantage in productivity over admins in a physical data center. (This may also be conservative as we’ve seen anecdotal evidence of a Ten to One advantage.) The cost of building that kind of automation (i.e. a private cloud) would be prohibitive for this small a data center.
Conclusions
The analysis shows that during the initial investment year one, AWS has a significant advantage of over 60 percent less that the colo approach. However, over the three year life span of a server AWS would costs almost 19 percent more than the colo approach. In year four, when a server refresh would be in order for the colo, the procurement cycle would repeat meaning a new investment of Three Million Dollars while the AWS approach cost would be steady state or declining slightly. Therefore, in year four the four year accumulated advantage swings to the favor of AWS by 18 percent.
It’s important to note that this model does not capture the value of time to market. The time required to build out each data center is dramatically different. While an AWS based datacenter of this size could be brought up for the first time in a matter of days (or hours), a physical datacenter of this size would take months to accomplish. In fact, companies that have used the cloud cite business agility as the number one advantage of a cloud over a physical data center – not cost differentials. Additionally, the time and cost required recreating the management tools, virtual machine support, monitoring and other software that would put a colo on par with a cloud for ease and rapidity of deployment (i.e. a private cloud) is considerable – perhaps more than the entire data center itself.
So what are your thoughts?
To download the white paper and supporting spreadsheet –
