March 04, 2009, 7:01 PM — In mid-February a group from UC Berkeley released a paper titled Above the Clouds: A Berkeley View of Cloud Computing that addresses the strengths and weaknesses of cloud computing.
I recommend firms considering moving some or all of their applications into the cloud take some time to read this paper. The paper is lengthy and not without its flaws, but it is a good resource for firms considering moving to cloud computing as well as for cloud computing vendors. It goes beyond the obvious value propositions of SaaS and cloud computing to look at business metrics to consider when evaluating these services.
Here are the cliff notes.
Three items regarding the economics of using SaaS are particularly interesting and worth spending some time weighing for your own particular situation. First is the cost of data transfer compared to other transfer methods. Second is the amount of computation the applications require. And finally, opportunity costs associated with taking advantage of leading edge technologies.
In addressing the data transfer issue, the paper uses an example of the time and cost associated with sending 10 Terabytes of data from its laboratory to Seattle, Washington. Like all good academic papers, the math formula is embedded, but the bottom line is that it would take more than 45 days to transfer the data and the network transfer fee would be $1000.
The suggested alternatives to using the Internet to transfer the data were to either ship ten 1 TB disks overnight or store the data in the cloud thereby removing the transfer bottleneck for any future uses. Shipping the disks incurred a one day transfer rate and fees in the $400 range, effectively saving time and money. The authors concluded it would be most cost effective to ship the disks, though they also wandered down the path of why WAN rates are so high and are not declining relative to other technologies. As an aside, they finger the high performance router as the culprit.
The section on application characteristics as a metric for determining the use of cloud computing was interesting. The point being, the more computation required, the greater the value of working in the cloud due to the vast numbers of CPU cycles available. The authors refer to this as cost associativity, meaning there is no cost penalty for using 20 times as much computing for 1/20th the time. The upside is that the computations get completed much more quickly than in a fixed environment such as your own data center with limited server availability.
Specific applications and processes called out by the authors as being especially good opportunities for cloud computing included mobile interactive applications, parallel batch processing applications, compute-intensive business analytics and decision support, symbolic mathematics, 3D animation, and long-term financial decision planning. Each had the caveat of the cost of data transmission, but considering the alternatives to WAN usage would mitigate that concern.
The third area of the paper of particular interest was the arguments for the use of cloud computing based on technology cycles. Technology trends show that within a typical 3-year depreciation lifecycle of a device both the hardware and software will have improved in performance and dropped in price by the end of the first year, essentially rendering two years in technology penalties and lost opportunities. Cloud computing vendors are able to offer new technologies and applications to customers providing them with immediate benefit.