Capacity Planning

Author: Thomas Bryant, Director of Advanced Technology & Products, Vizioncore

One of the most talked about areas in systems monitoring these days seems to be capacity planning; and it’s one that’s heavily clouded. The main reason is that every environment is different. Even though we all tend to run similar application stacks, how we use them is unique to our individual behaviors and business requirements.

In most cases this causes us to look for tools that can develop simple models, with virtualization being put into those models. Recently I spoke with several SMB & Enterprise level users and they told me that they’re essentially looking for a way to take a ‘slice’, and to know how many of those slices they still have and when they’ll run out. This approach assumes 100% utilization of the resources, and that’s just not realistic.

The problem is that people are still thinking about capacity from a physical hardware perspective where you get X resources because it may need X resources. With all the benefits that come with virtualization it’s time for us as a community to develop new models that can adapt to each environment, and better yet accurately predict performance and capacity bottlenecks. This is why I’m a strong proponent of looking at each individual workload and correlating those workloads vs. ‘servers or vms’.

A virtual machine is nothing more than a shim, a place to wedge an application that performs some task. If you start to look at the VM as nothing more than an application workload, that will model far better to true capacity planning for a specific resource. Does 1 vCPU for a file server equal 1 vCPU for a SQL server? Certainly not, but the general capacity models don’t take that into account. Likewise disk utilization on a file server is going to be far higher than that of a print server. So looking at the application stacks or workloads is required if you want any level of clarity. In the end this will lead to better utilization of resources, and best of all it can help predict when new hardware will be required for net-new capacity with a relatively high level of accuracy.

-Thomas Bryant