When is Virtualization Beneficial?

There have been many discussions surrounding the various different virtualization technologies and the merits and downfalls of them. Virtualization is not new, we all know that, but every now and again it turns up in the top 3 list of IT buzz words and then silently fades away. To get an understanding as to how important Redhat thought it was two years ago, they delayed the launch of RHEL5 by some number of months so that they could have an 'Enterprise Virtualization' offering. Just out of my own curiosity, I would love to know what the adoption rate of virtualization on RHEL5 platforms are - My guess is that it would not even 10-15% of Redhat's total RHEL5 subscriptions.

There have been many discussions surrounding the various different virtualization technologies and the merits and downfalls of them. Virtualization is not new, we all know that, but every now and again it turns up in the top 3 list of IT buzz words and then silently fades away. To get an understanding as to how important Redhat thought it was two years ago, they delayed the launch of RHEL5 by some number of months so that they could have an 'Enterprise Virtualization' offering. Just out of my own curiosity, I would love to know what the adoption rate of virtualization on RHEL5 platforms are - My guess is that it would not even 10-15% of Redhat's total RHEL5 subscriptions.

So why do people get all excited and foamy at the mouth over virtual machines? I could be wrong here but I am guessing that the majority of virtualization adoption would be to provide an extra machine that does not require too much resources to run a particular OS or service. The primary example is running a windows guest OS on a linux machine. This is usually done when the adopter requires a service that can only run on a particular OS.

This is very useful for obvious reasons and provides an excellent and tidy solution to a messy problem - two machines, screens, infrastructure etc ... But where does virtualization fit in in the broader enterprise type environments? Virtualization providers will dedicated millions of pixels to tell you about how virtualization is the way forward and that it will save money and increase up time and provide better efficiency in the datacentre and so on and so forth. To an extent, this is very true. The capabilities of the tools from say, VMware, can provide are phenomenal. To give a brief indication, the vmotion technology and data centre manager from VMware allows complete automation of the migration of a virtual guest from one physical server to another and then power down the machine that no longer needs to be on. And then power the machine backup and move the guest OS back when the need arises dependent on configured factors such as increase in load. It is easy to see how, in large scale environments, this technology like this can be beneficial.

Unfortunately, there are two factors in the above scenario which can put the dampers on it a little bit. First, proprietary virtualization technology providers 's data centre and virtualization management tools are usually very costly almost to the point where the savings gained from efficient management of the data center or hardware resources are lost when offset against the cost of the technology in the first place. Secondly, the management of such architectures which are not greenfields setup can be tricky and potentially risky to business. And it is for these two reasons that I personally have not been a full adopter of virtualization. I have found myself more often than not falling into the category of the user who needs an extra box and finds the convenience of 'xm create windoz3'

However, a project that I have been working on lately has changed my view on virtualization significantly - to the point that I believe EVERYTHING must be virtualized. The conundrum that was faced was a requirement for 4 HA clusters - oh and also that there was no money for fancy virtualization software. In traditional setups, this would necessitate 8 servers - a luxury that was not afforded to us. So the idea was to implement 5 physical servers and run 8 virtual machines - two on each physical host and one physical servers sitting twiddling it's thumbs until something goes wrong.

The strategy was to use a clustered file system for the 5 physical machines and have the image of the guests reside on the GFS. That way migrating a guest from one physical machine to another was made significantly simpler. This also reduced our infrastructure requirements in terms of switchports, rackspace, power and actual hardware. We believe operating our clusters this way has improved our efficiency from 50% to 80%, purely in terms of hardware required. This example only starts to show what can be done with virtualization. By adding more physical nodes to the GFS, it is possible to operate more and more VMs and thus increase the efficiency further.

The above example, for our environment, was best achieved using RHEL, but it matters not with what linux distro is used. The point was that it can be done and it was done using open source software.