Virtualisation technology has greatly enhanced business agility, enabling IT departments to quickly meet the computational demands of various business units. If sales and marketing wanted additional servers, IT could easily create new VMs and deploy those servers; even within a couple of hours if needed. Unfortunately, the ease of spinning up new servers combined with certain misconceptions is often leading to VM sprawl.
If left unchecked, VM sprawl can rapidly deplete computing resources, preventing some business units from deploying new servers even if those resources are actually underutilised.
One resource whose underutilisation often goes unnoticed is CPU. Normally, when we’re given unrestricted access to a pool of CPUs, most of us have this natural instinct to allocate as many CPUs to an application as possible.
Theoretically, it should be possible for today’s multi-threaded applications to utilise several CPUs. But in reality, such applications aren’t as common as we think. As a result, an application allocated with a certain number of CPUs may only use up some CPU cores and leave the rest untouched.
Another resource that’s often unintentionally underutilised is memory. Unlike CPUs, memory resources are more likely to be “consumed” by applications. That’s because an application will gradually put as much into memory as it can to allow faster retrieval later on. The thing is, not everything that’s placed into memory is actively used. They’re only there just in case they’ll be needed. So even if you see 100% utilisation, it’s possible that only half of that is actively used.
Lastly, we have this notion that virtual machines are free. They’re not. Whenever you spin them up, they start consuming storage space, memory, CPU, and even network resources. Chargeback and showback tools are actually able to show where resources are going, making it easier to associate consumption to a corresponding cost.
If we can just eliminate these misconceptions of CPU and memory consumption as well as of the so-called “free virtual machine”, we would be in a better position to manage our virtualisation infrastructures more efficiently.Back to Top