Excessively fat virtual worlds – come on, it's your guilty secret

Does my server look big in this?

Excessive VM embiggening issues

On the other side of the equation, giving a virtual machine too much capacity can negatively impact performance. This is due to the fact a virtual machine has to take into effect managing and scheduling between all the virtual CPUs it has.

This can introduce an overhead greater than the extra resources gained by adding CPUs.

Often, customers ask for additional CPUs to improve performance further and are then disappointed when the performance of the virtual machines decreases.

For each and every virtual guest, every CPU and Gigabyte of RAM adds an amount of overhead to the physical host; even when powered off. It would be easy enough to resize the odd machine that is over-specified on an ad hoc basis.

Using that approach is just not feasible in large estates where tens or even hundreds of machines get deployed on a daily basis. It becomes that much harder to monitor the machines and resize. So how does the unfortunate administrator deal with this issue?

The approach is twofold: carrot and stick. Any large company worth its salt should be using a charging model. The more resources a business unit uses or otherwise claims, the more they pay.

On the one side, reclaimed resources mean a reduction in charges per month to the customers, but by the same token, businesses are skeptical of reducing the costs but maintaining the current service levels. Getting the business on board as the first step is absolutely critical.

All this leads onto the inevitable question of what is the best way for an administrator to start reclaiming under-utilised capacity. The first step is proving the wastage in black and white. There are a number of ways to do this.

The old fashioned way involves looking at metrics manually, which could potentially cost a small fortune when scaled up. The smart money is on tools specially designed to do the job, that are not only virtual machine-aware but incorporate intelligent tuning algorithms that will advise sensible reductions.

Such tools include VMTurbo (free edition available) and VMware vCOPs.

Be warned that collecting the metrics may take a while. The longer the collection duration, the better. The data collection period needs to be long enough so that any usual compute peaks can be taken into account.

Once you have the cold, hard facts, it comes down to doing the sums of calculating the potential cost savings based on the new sizing and discussing them with the owners of the servers – be that your manager or a business representative.