July 2009 Archives

Author: Daniel Beveridge, Director of Virtualization Strategy - VIRTERA

Central to the promise of virtualization is the ability to decouple workloads from their underlying layers of technology. In the case of hypervisors, this means that the OS can have a common set of drivers while freely moving between hardware platforms. For application virtualization, the same mobility allows dynamic delivery and execution of software across a range of operating system versions. Major vendors such as VMware actively market virtualization as a way to avoid vendor lock-in at the server level. Companies have welcomed the new freedom, reveling in their new found ability to negotiate more aggressively with hardware vendors who must now account for the ease of migration to new hardware platforms.

The virtualization market has matured over the last few years and the freedoms associated with virtualization of the technology stack have expanded and increasingly become a requirement for companies looking to make investments in the datacenter. What was previously just a possibility has rapidly become a necessity. Ironically, major virtualization vendors have not fully grasped the implications of the revolution they unleashed. Product strategy lags customer expectation by continuing to bundle products in a manner out of step with the elimination of vendor lock-in promised. Companies feel cornered into virtualization investments that may free them from physical server lock-in only to encumber them with hypervisor and associated technology stack lock-in. 

Even after working on green IT projects for the past four years, I keep being amazed how one's intuition is insufficient to fix data center air flow and cooling issues.  I'm reminded that data always needs to be collected and reviewed in order to fix problems.

Recently, I was asked to take a look at a small 1,800 square foot data center running at about 100 watts per square foot.  It contains 29 servers and 6 computer room air conditioners (CRAC).  As data centers go, this is pretty simple.

The problem was that about 20% of the data center is running hot - hot being better than 90 degrees F and 99 degrees F in a few places.  I was asked how many new CRACs should be installed, and where should they be placed.  The question on the surface certainly appeared reasonable - even to me.

I collected information about the data center's environment and reviewed a air flow and temperature model which I sat down and studied.  Nothing really jumped out at me.  The servers running hot were 18 feet in front of a CRAC with nothing between the two.

As an ex-CTO of a global Financial Services company in New York City, I worked through Y2K, the dot-com bubble, and the emergence of x86-based server virtualization.  Our first introduction to server based virtualization occurred in early 2001 after consolidating several data centers into a single global data center. As a result of the consolidations, we were in need of finding ways to reduce the environmental impact that the consolidations introduced or face expansion of the production data center. To address these environmental issues we adopted a virtualization strategy for our test / dev environments late that same year.

While not highly scalable, the benefits we realized were fairly significant and based on our early successes, we quickly evaluated data center ready hypervisors when they were introduced in late 2003/4. We became the first financial institution to leverage virtualization in a production data center in early 2005.