Is your server farm out of control? Is it characterised by a sprawling, disorganised pool of processing resource in which individual boxes run at rates far below their potential capacity?

If so - and you are by no means alone - then it may be time for a server farm redesign, using virtualisation technology. This will allow your company to migrate processing work onto fewer servers, which in turn will reduce the administrative burden placed on systems administrators and cut down on the power and cooling you need to keep your server farm up and running.

That may sound highly attractive -- but of course, there's a caveat: an ill-planned, over-zealous virtualisation strategy may result in an environment that proves insufficient when traffic volumes soar or individual servers fail, says Jan Zelezinski, strategic architect at Logicalis.

What's needed, he says, is a thorough capacity-planning exercise that will ensure that your company has sufficient hardware and software resources to support its planned server farm migration. If this is not done correctly, then the solution may not provide the expected performance and service levels.

"Server consolidation starts with capacity planning to understand the current inventory of servers, their workload, and their utilisation," he explains. In most cases, automated capacity planning tools are used to collect data about peaks and troughs in server utilisation by applications. From there, it is possible to assess the current server landscape, identify opportunities for consolidation and model a realistic consolidation plan.

A comprehensive capacity planning exercise should:

1. Provide an analysis of server utilisation;
2. Identify which servers are candidates for redundancy;
3. Determine the suitability of applications for virtualisation;
4. Estimate potential return on investment from energy savings.

Your Comments and Questions

Mandy Shaw, 8 months ago

To add to Jan's point, in a Unix environment it is even possible, with such dynamic logical partitioning setups, to have processors left on standby in the system, activated (manually or automatically) at peak times only, and paid for only when required. I can't personally think of a retail customer that is using an Intel platform for its line-of-business applications - any counter-examples gratefully accepted!

Jan Zelezinski, 8 months ago

In the Intel space, applications that display very high peaks in demand for compute resources may not be right for virtualisation as there are limits to the amount of resources that can be given to any one VM. Part of the CP exercise is to establish anomalies like this. However there are alternative technologies such as DLPAR (Dynamic Logical Partitions) which came from the mainframe world and are now prevalent in both highend & midrange UNIX systems and are more suited to high transational workloads of typical retail apps. This in itself is a form of vitualisation as multiple LPARs can be created on multi CPU H/W partitions which can be dynamically changed depending on workloads.

Amanda Smith, 8 months ago

Can you apply this exercise to extremely unusual peaks in demand too, or does that require additional planning/resources? I'm thinking of retailers, for instance, which might see a huge spike in demand in the days running up to Christmas, which is very different to the daily spikes in demand caused by people logging on at 9am and trying to access the same applications.

Jan Zelezinski, 8 months ago

Not necessarily. During the capacity planning process data is collected over a period of 30+ days or more if required. From this data a virtualisation solution can be designed to handle peaks in demand by providing a "pool" of physical resources that can manage fluctuations in workloads and provide HA.

George Black, 8 months ago

There seems to be a bit of a 'trade-off' here. Is that the case? And how does one best balance requirements in terms of the efficient supply of 'day-to-day' demand and providing enough spare capacity to deal with unusual peaks in demand?

What's Your Question or Comment?

Name

Email Address

Your Website URL (please include http://)

Your Question or Comment

type the text from the image