I sometimes wonder how many people realise that having a virtual datacentre is a balancing act between staying fit and agile or becoming over weight and cumbersome.
I see it time and time again where people think that the only use for virtualising servers is to consolidate physical boxes. They take a look at their physical box and say it has 2gb of ram, 2.0 dual core processor and 128gb mirrored system drive. So when they virtualise it what do they do…. Obvious, isnt it?
They monitor the system, look at its usage over a few days and work out its peaks and dips in system resource usage. Take some averages and then P2V and give it only what it needs and a little extra for good house keeping in the virtual environment.
Well you would be wrong as I see it all the time, they give the box 2gb of dedicated RAM, they create a single 128gb VHD and then assign dual virtual processors. Then over time they oversubscribed and run low on resource on the host server. To ease this scenario a new host is added and then more machines can be virtualised, before you know it your virtual infrastructure is bloating with hosts. Then as people become familiar they realise how easy it is to add a new guest and before you know it all hell is breaking loose and the data centre is falling on its knees.
In order to keep the datacentre agile you have to employ a great deal of change control, guest life cycle control, monitoring of the hosts, best practice and applications for sizing and capacity.
Best practice with VMware for example will tell you that most guests do not benefit from more than one vCPU due to how the HEC (Hardware Execution Cycles) are managed. In fact in my time I have only come across a couple of servers where I have proved that an additional vCPU was required. I tested CPU affinity but found it was not required.
Change control allows for some form of record of systems and why they were added and also gives a degree of control as to what happens in the environment. With the advent of Cloud and the versatile platforms a lot of change control will be automated using products like System Centre Opallis or VMware orhestrator as end users provision their own machines.
There a vast number of third party tools available for capacity and sizing, my experience lies with the VMware offerings like capacity planner. These allow you to accurately monitor and size for physical machines and help to maintain your virtual hosts.
Life cycle management is a new one for me, it’s basically a cloud feature and it allows datacentres to set life times for guests and remove them when their life time has expired. As I find out more I will post up about this.