By Lew Moorman, President, Cloud & Chief Strategy Officer, Rackspace Hosting
By now, anyone serious about cloud computing has read (or read about) thethat aims to debunk the revolution. The cloud is not ready, too expensive, and easy to replicate, they claim.
First off, it is hard to ignore McKinsey taking such a firm position (full disclosure: I am an alumnus of McKinsey). Second, the hype of cloud computing has grown so loud that a little water to douse the flames is not unexpected.
The report has some good thoughts in it, but overall, they are missing the bigger point and underestimating the benefits of cloud computing. That said, the report raises their credentials in the space and puts into question the rise of cloud computing. However, I fundamentally challenge the assertion that cloud computing is uneconomic.
More to say on that, but first here are some high points from the report:
We Need A Clear Definition
McKinsey makes an urgent plea for clarity and takes a simple approach. Clouds have 3 characteristics, they claim:
- Abstracted hardware
- Purchased as subscription (opex over capex)
- Elasticity – easy up/down
I think they have this essentially right. We use 3 slightly different characteristics:
- Pooled computing
- Powered by software
- Delivered over the web
They focus on benefits; we focus on the underlying technical cause of the benefits. Pooled is essential to abstracting and describes the more important point — you are pooling users on resources to get higher utilization. I am not sure the subscription point is critical. If you want to pre-purchase or effectively buy resources, I think this will be available over time (plus, non-cloud hosting is bought as a subscription too). The bigger point is elasticity. We think this is part of the “powered by software” piece of our definition. Pooled resources divided by software in an automated way means you can go up/down and pay for what you use. It also means lower opex since most core functions are automated. Finally, we say it must be over the web. This means it is not on your LAN and includes some concept of remote resources. Does that mean there cannot be private clouds? Well, that is another post.
Clouds vs. Cloud Services
I like how McKinsey makes a distinction between infrastructure and applications. They say Amazon and AppEngine are clouds (as would be The Rackspace Cloud). Salesforce and Gmail are Cloud services (as is our Mailtrust offer). Most of the world has moved to IaaS, PaaS, and SaaS definitions to draw distinctions. McKinsey just combined the first two. The important point is there are flavors and we need to be clear about them. Not all clouds can be used to solve all problems and the distinction between bringing your own code (cloud) and consuming an application (cloud services) is a worthwhile division.
The main piece of the report centers on costing Cloud vs. DIY. They suggest that effectively cloud computing costs 5-7x DIY from a hardware/DC perspective and only creates a 10% reduction in labor.
The report is clearly a back-of-the-envelope view (honestly the data is hard to analyze because there are many inconsistencies and odd unexplained assumptions) and meant to shock. It does. But, is it right? Here are some key issues I have with the report:
$43/Month For An In-House CPU
Their own math does not even work to prove this assumption. They say a 2 CPU server costs $14,000. Amortized over 36 months that is $388 a month or $194 per CPU. Now honestly, who pays $14,000 for a windows server? Cut it in half, that is still $97 per month. Good DC space costs $1200/foot. Over 10 years that is another $10/month. This does not include any people, licenses (OS and virtualization), power or maintenance – none of which is insignificant.
10% Reduction In Labor
The labor analysis includes all the IT application functions. Yes, you still need help desk support, development, telecom people, etc. But, where are the DC engineers in their analysis? And the network administrators? The smart hands? The virtualization team? The procurement and inventory people? They are not mentioned. But these resources are where you get the real savings. Furthermore, if you use some of the emerging automation built for the cloud (e.g. Rightscale or Cloudkick) reductions in administrators should go down even more.
Benefits of Cloud
While the “promise” of the cloud is outlined, the benefits have no role in the economic analysis. How much is saved by being able to pay only when you use a server? You can’t turn a server off that you purchased from Dell. What value is there in knowing you can scale at a moments notice with no need for inventory or project resources? What value is there in being able to get any OS or image loaded whenever you need it? What value is there in having a team of experts on call at a service provider should you run into trouble?
The Bigger Question: Core Value
I am sure there are plenty of comebacks to all my points. But, here is the bigger question that should not be lost on McKinsey:
Does it make sense for non-IT companies to be outstanding at computing operations?
Does putting 100’s of millions of dollars into DCs and people to support servers have anything to do with the core mission of most companies?
This is the biggest value of cloud computing and the reason it is a revolution. The ability to consume it effectively as a service means you can offload it and focus on your core.
I hold the hype is real because of this opportunity for focus plus it is radically cheaper than most companies can do it themselves.
Currently businesses spend 75% of their IT budget on maintenance, 25% on strategic activities. The real promise of cloud computing is inverting that equation. Time will tell if it lives up to the hype.
Follow Lew Moorman on Twitter: @lewmoorman