Comment on the McKinsey Cloud Computing Report
There is a bit of buzz on some of the Google Cloud Computing groups about William Forrest’s entitled “Clearing the air on cloud Computing”, March 2009. I’ve taken a look at the presentation and while some of his analysis is valid, there is a fair amount that tends to miss the point of Utility Computing. I plan to write up a full analysis but I just could not wait and decided to give you a couple of highlights of my thinking about the report.
First, McKinsey is great at creating buzz. I am sure this drives business to them. Many in IT use McKinsey and others reports for analysis gathering and in general these reports do raise important questions about the applicability of a particular technology. “Clearing the air…” starts off focusing on all of the buzz that Cloud Computing has generated over the last 2 - 3 years. As Forrest states, “Cloud Computing is approaching the top of the Gartner Hype-cycle”.
I agree with him here. There is a lot of noise about Cloud computing. However, I see this as a good thing. Cloud computing has given those that practice the art of system and program design a whole new set of tools. Tools, that allow us to more easily deploy system that are inherently scalable, elastic and fault resistant. I won’t go into those here but think about this. Animoto could never have delivered a system capable of utilizing 3000+ servers on a shoe string budget just 3 years ago and especially in the short time frame they were able to stand those “servers” up. I don’t care how much money your company has to spend on IT, ordering 3000 servers, getting them racked, powered, cooled, named and finding room for all the little manuals that talk about how to lift the server out of its box would be a daunting task. Suffice it to say, Cloud Computing Hype is in overdrive because people are realizing the power the Cloud toolbox provides. Forrest doesn’t say much on this aspect of Cloud.
On the definition front, Forrest makes claims that a common definition does not exist. He quotes from Gartner, IDC, the Financial Times, Merrill Lynch and others as examples of the confusion out there. Um…. excuse me. But the last time I needed a definition on a particular technology I didn’t call my broker at Merrill Lynch for a definition. I might check IBM or even Wikipedia for a definition.
In fact, the proposed McKinsey - Forrest definition for cloud computing does not cover the full genre of Cloud. Forrest only discusses one of the 3 tiers of cloud computing; Infrastructure as a Service (IaaS). He further makes the mistake of classifying Google App Engine as an IaaS. The App Engine is a great platform to develop and deliver applications from. Google App Engine is a prime example of the Platform as a Service (PaaS) genre of Cloud Computing. It is not an Infrastructure environment. A developer will use PaaS to develop and deliver software solutions. They do not have direct access to the infrastructure, nor do they need to. If you need IaaS then companies like Amazon Web Services, 3Tera and others are the ones to look at. If anything, Forrest has contributed to the confusion as to what Cloud Computing is instead of clearing the air as he insists is required.
Finally, and this is the bit that really stumped me is his cost analysis. I can arrive at a CPU equivalent for a purchased server of around $45 / month. I have to make some assumptions such as depreciating the asset over 3 years and including power costs. However, to hit this number, I have to neglect other facility charges, property tax (for states that charge property tax on capital) and maintenance contracts. What I can not arrive at is his numbers for Amazon EC2 equivalent CPU costs. He shows values ranging from about $75 - $180 / month per CPU equivalent which by my calculations is baffling at the top end.
Amazon EC2 has a very simple pricing model. There are two cost schedules. A reserved and on-demand pricing model. The On-Demand model is the one I show here. It is the more expensive of the two.
| Size | Cores | $/Hr | $/Day | $/Month | $/Core / Month |
| Small | 1 | $0.10 | $2.40 | $73.00 | $73.00 |
| Large | 4 | $0.40 | $9.60 | $292.00 | $73.00 |
| Extra Large | 8 | $0.80 | $19.20 | $584.00 | $73.00 |
| High-CPU Medium | 2 | $0.20 | $4.80 | $146.00 | $73.00 |
| High-CPU Extra Large | 8 | $0.80 | $19.20 | $584.00 | $73.00 |
By my calculations, the monthly per core cost is always $73 per month for a non-Windows server. There is no costs for power, property tax and maintenance. More importantly, if I have architected my system to take advantage of Amazon’s services, I can scale these machines based on demand. Scale up and down. For me that is the significant value proposition that the McKinsey report fails to mention. That is one of the main reasons people are so excited about Cloud Computing. That is a new tool for us to architect cost effective, energy efficient infrastructure unlike any that we have been able to design and build before.
Quello è tutto per oggi!
-Paul Laskin
