Brian Gammage

For most organizations, demand for computing capability grows faster than revenues and can continue to grow even if revenues decline. Data center capacity often restricts the scope to meet this demand through traditional approaches. Over recent years, vendors have created compelling business cases for virtualization technologies that address this issue by increasing the utilization of servers. With IT budgets under pressure, the marketing of these is getting more pointed – in some cases suggesting virtualizing existing hardware is always the lowest cost route to additional capacity.

Virtualization undoutedly delivers utilization improvements, but marketing messages like this tend to ignore the impact of Moore’s Law on the performance of hardware. Moore’s Law underpins sustained gains in server density and performance, at reducing price points – we get more capability with new hardware at the same price point as the old and so next year’s server does more work than last year’s. When it comes to using virtualization to gain additional effective capacity, that matters.

Consider this hypothetical simple example based on a relatively low cost “volume” server – with two quad-core, x86 processors. Let’s assume that a 2008 server can support 4 virtual machines, a 2009 server 8 and a 2010 server will support 16. These are exaggerated differences, but the numbers make the math simple – the trend remains valid. A company looking to support 14 new “virtualizable” workloads in its data center can:

1. Virtualize five 2008 servers by buying five copies of the virtualization software
2. Buy two 2009 servers and two copies of the virtualization software
3. Wait until 2010 and buy one server with a single copy of the virtualization software

Lets compare the cost of options 1 and 3. Option 1 is only the lower cost option if the server costs more than four times as much as the virtualization software. A quick scan of the web sites of the various vendors involved suggests this is not the case - the virtualization software and servers typically have prices of the same order of magnitude.

What does this tell us?

• Moore’s Law drives increasing marginal returns from virtualization on new servers.
• Returns from virtualization on existing servers diminish with increasing age of the server.

Perhaps, more interesting to those organizations with an urgent need to conserve cash: 

• Increasing returns may justify postponing investments in virtualization and server hardware. 

This type of “financial view”  is going to become increasingly important as we look to the cloud and to external providers as the source of additional data center capacity. If the cost of creating additional capacity “in house” falls each year, the pricing and review periods for contracts will need to be structured accordingly.