Recently I have been working with a global organisation to assist with capacity planning and wastage of their virtual estate. This organisation has a significant number of business units and network zones that need to have their capacity planning performed independently, but the clusters within those boundaries are generally managed by the same vCenter server with geographically focussed datacenters. The challenge therefore was how do we manage capacity of our groups of clusters (this customer has hundreds of clusters)? One thing to note is that this customer has a standard (which is then reflected in a vROps policy) to only use an allocation based model of capacity utilisation and they were unable, for a variety of reasons, to move to a demand based model.
In vROps we have two means of grouping objects; custom groups and custom datacenters. Unfortunately neither of these is ideal for our purposes, due to the reasons shown in the table below:
When discussing these with the customer they asked the obvious question:
Is there a way to determine the time remaining of a group of clusters based on the time remaining of the individual clusters via a super metric?
The simple answer is no, and the more complex answer is still no
Take the following simple example where we have two identically sized clusters and every VM that we deploy is the same size and we deploy a set number of VMs per day (to make this example easier to understand)
- Cluster A has capacity for 10 VMS and we deploy 2 VMs per day
- Cluster B has capacity for 20 VMs and we deploy 1 VM per day
Based on the above the time remaining for each of these would be as follows:
- Cluster A – 5 days
- Cluster B – 20 days
So what is the time remaining for the clusters as a group? 20 days? 25 days? Something else? The table below shows us how the group of clusters will be filled and therefore how much time remaining there is:
|Cluster A||Capacity Remaining||10||8||6||4||2||0|
|Cluster B||Capacity Remaining||20||19||18||17||16||15||12||9||6||3||0|
This is a a simple example and yet the maths is pretty complex, imagine this with 10 clusters, differing virtual machine sizes, different and changeable deployment patterns and the ability to define a single value via the super metric mechanism is not suitable.
The answer therefore seems to be custom datacenters, as by using these we gain access to vROps inbuilt capacity engine to calculate the time remaining figure for groups of clusters. Custom Datacenters are top-level objects within vROps and every benefit that goes with that.
There are still some challenges that I’ve experienced specifically around consistency of numbers whereby the number of virtual machines remaining in the clusters that make up a custom datacenter don’t necessarily add up to the number of VMs remaining for the custom datacenter itself. This can however be masked by the use of super metrics.
The second and much larger challenge however is that there is no dynamic membership mechanism for custom datacenters. In smaller environment this may not be so much of a problem whereby the membership can be managed manually. However, in environments with hundreds of clusters, then managing this manually via the UI is both impractical and prone to mistakes.
With my current customer we’ve used the vROps API both to create the custom datacenters but also to manage the relationships between the custom datacenters and the underlying clusters. More information on how to use the vROps API can be found at the following links:
In doing so PowervROps was born, a module that allows the use of the vROps API directly from within PowerShell, but that is worthy of a post all to itself: