Dave Pearce asks the pIT stop panel:

When should you virtualise a server and when should you not? What are the most important aspects of the decision process?

The pIT stop panel's replies:

Virtualisation is one of the most significant technologies affecting IT departments at the moment. Companies are virtualising their environments for many reasons. For some the ability to consolidate servers, improve asset utilisation and save on energy costs is the key. For others it’s the ability to have a more flexible IT estate, with faster development and testing cycles – together with more flexible and cost effective business continuity arrangements.

We generally recommend that companies look at virtualisation from three perspectives when deciding what environments to virtualise: technical, management, and commercial.

On the technical side our normal recommendation is that the decision whether to virtualise an environment, or not, needs to be based on benchmarking of the actual applications involved.

On the management side many customers are finding their overall management costs, whether direct or through a service provider, are not actually decreasing in the way they have thought. This is because the physical complexity that generated a management burden has been replaced with a virtualised complexity of the same magnitude. Anyone considering virtualisation needs also to look at how the management framework can be simplified, through stack standardisation, management automation, and an overall simplification of the current environment. Moving from physical to virtual will produce benefits but nowhere near as much as when simplification, standardisation, and automation are also included in the picture. These tasks generally need to happen before the implementation of any virtualisation solution, in order to generate more benefit.

One of the areas to watch out for on the commercial side concerns software vendor support. Before virtualising any application the commercial team in the enterprise needs to check whether any of the existing support and/or service agreements would be affected. Many application vendors will only provide support when their applications are running on a physical environment, and will force the customer to replicate the problem on a physical environment before providing support. This can increase the risk and decrease the benefits that virtualisation is able to deliver.

By David Mitchell, senior vice president of IT research, Ovum

This is a great question and many facets of an IT environment can shape a decision on how to implement virtualisation. In general as a 30,000ft view, applications running on servers that are two years old or more are the most suitable candidates. Moving applications from hardware of this age onto currently available multi-core/multi-processor servers will provide similar or even superior performance than when hosted on their original physical hardware. Applications such as heavily used databases with a large compute and disk I/O are less suitable to virtualisation.

Steps to a successful virtualisation implementation:
Moving to a virtualised environment should be considered a process with the two most important areas being an accurate data centre environmental survey which shows actual levels of utilisation, and detailed planning.

Environmental survey
The environmental survey will reveal the most suitable systems to virtualise - just because a brand new multi-core server has been deployed for an application doesn’t mean a given application is actually using the full resources available to it. A survey also provides data to accurately estimate the virtual resources required for carrying out a virtualisation effort, so hardware can be correctly sized to meet your needs today and tomorrow.

Detailed planning
Use the data from the environmental survey to actively end the life of applica tions that are no longer needed. It should also be used to determine the following - estimated virtual host hardware requirements and the number of hosts needed; the method of migrating servers/applications onto the virtual platform; the potential consolidation ratio (number of virtual machines per host). In addition to this, the environmental survey will provide the data needed to create a migration strategy for implementation.

Suitability
There are certain high-level rules-of-thumb that you can use to estimate the suitability of applications to be hosted in a virtualised environment. Less suitable systems include those with a high memory requirement - while not a limiting factor in itself, it will significantly limit the consolidation ratio and provide a lower return on investment.

Also less suitable are systems that have a very high disk I/O requirement. Disk I/O can be a bottleneck in virtualised platforms although it should be noted that new virtualisation assist technologies and faster storage make this less of an issue. Any application with a moderate CPU/network/disk or memory requirement, particularly if the hardware on which they are based is two years old or older, will yield the best consolidation ratios. It may be possible in these cases to achieve a consolidation ratio of 16:1 or higher on a high-spec, dual-processor, multi-core virtual host with 16-32GB of RAM installed.

Business Criticality
One important factor to consider in terms of the order in which applications should be virtualised is their importance to the business. It is important to determine the business criticality of your applications. Those that are most critical should be the last applications to be brought into the virtualised environment. It is only once you have confidence in, and good knowledge of, the tool sets, the processes, and most importantly the skill sets within the organisation, should you think about bringing in those most business critical applications into the virtualised environment.

Here is a link to an IT@intel virtualisation whitepaper that Bill Sunderland and I co-authored around virtualisation implementation in Intel:
http://www.intel.com/it/pdf/Implementing_Virtualization_in_a_Global_Business-Computing_Environment.pdf

By Stephen Anderson, Intel

The first and perhaps most obvious question is - how efficient is your existing server estate? If a server is already running at high utilisation rate perhaps you should leave it alone.

Virtualisation makes most sense when it's part of a broader consolidation and/or service management strategy. Rather than simply considering individual servers, pools of servers can support new processes, cutting across the current boundaries of the IT organisation. Thus, for example, servers used for quality assurance (QA) or testing new applications before rollout, are only used some of the time - they are by nature candidates for virtualisation. Software developers regularly need to target different system configurations as part of the development process. Development servers are thus also a prime candidate for virtualisation. Indeed, VMware's initial period of serious growth came on the back of development and QA consolidation, before Microsoft Windows was even supported as a production environment.

Virtualisation should be considered as a core element in server manageability, but is far from being a sufficient condition for it.

One of the emerging use cases for virtualisation is disaster recovery, where if one server drops, another can take up the strain. In this case, virtualisation makes sense even if the server is already running efficiently.

All in all, the question increasingly becomes why should a server not be virtualised? Not only does virtualisation offer manageability benefits, but it's also now a given - Windows Server 2008 is shipping with a hypervisor, and VMware is already well entrenched in many IT departments. All Unix servers now ship with virtualisation as standard. Some organisations are now consolidating distributed Linux workloads onto IBM mainframes.

With that in mind, some things to look out for - Don't virtualise:

• If the server is running an I/O intensive workload, such as a database
• If you don't have the skills in place to take advantage
• If you're not ready for intensive software licensing discussions

James Governor, co-founder and principal analyst, RedMonk

Increasingly the question to ask is – why should we not virtualise a server, rather than when should we? There are of course a variety of technical considerations, but those are well discussed by our panel of experts above.

The business drivers for virtualisation are growing – cost reduction, IT efficiency, energy costs, environmentalism, agility, business continuity, time to market, manageability – the list goes on.

It should not be difficult for any IT manager to make a coherent and compelling business case for virtualisation to their chief executive. But when you do – leave the technical talk to an absolute minimum. I’ve heard too many cases of people saying they can’t get boardroom backing because the directors don’t understand the concept. Don’t try to educate them – just explain the benefits.

For many organisations, the imminent release of Microsoft’s Hyper-V virtualisation product for Windows Server 2008 will be the catalyst for a serious evaluation of the technology. Like it or not, Microsoft’s entry into a market such as this brings it into the mainstream. Once servers start shipping with built-in virtualisation capability – even if you need a licence key to switch it on – it becomes an easier decision. Market leader VMware will benefit too as companies that previously thought virtualisation too technical, expensive, or complex will be led to consider the technology through the sheer force of Microsoft marketing.

The question of whether or not to virtualise a particular server is stage two – ably assisted by the expert advice provided above. The first question of whether or not to investigate the potential of virtualisation has an easy answer: yes.

By Bryan Glick, editor, Computing

Read more about the pIT stop here: www.computing.co.uk/pitstop