Insights

One of the drivers for capacity planning in on-premise datacentres is the lead time required to provision new infrastructure and the requirement to ensure there is room and power to accommodate it.

When the concept of Cloud computing first started to become mainstream in the mid-2000 the death knell for capacity planning was sounded, just as it had been when virtual servers started to come to prominence a few years earlier.

In truth as it becomes easier to provision infrastructure the more capacity planning is required. The difference is the way you do it, not that it is no longer needed.

Two of the main advantages of the Cloud are:

1. It is very quick and simple to provision the infrastructure you require, such as servers, storage and network capacity
2. There are many ways in which these can be provisioned at different price points to suit every budget

The problem with this is that unless managed very carefully, it is all too easy to over-provision infrastructure and not provision it in the most cost-effective way to meet your business needs.

In order to manage the use of capacity effectively, it is still important to determine what the current use is, what the change in use is (up or down), when the peak periods happen (daily, weekly, monthly, annual) and to add in the business and application changes that will be happening along with their estimated usage.

Once this is understood you then need to determine the best way to provision the infrastructure, e.g. for compute; On-demand, Burstable, Reserved Instances, Savings Plans, Spot Instances, Clustered or Serverless. With similar multiple choices to be made for storage and network provision.

So far, so normal capacity planning. The next stage is to establish what the charges will be made by your cloud provider/s on an annual, monthly, weekly and ideally daily basis. This can be very tricky to determine, partly due to the complex charging policies that differ from region to region and partly due to the way the charges are applied, usually a charge for the infrastructure itself but also usage charges for such things as bytes transferred, bytes stored, CPU used above a threshold and inter-region or zone network costs often with a free tier up to a certain amount.

In effect, there are two capacity plans that need to be created, one for the infrastructure and one for the way it is going to be charged for.

The big difference is that your capacity planning now needs to track against the predicted cost rather than the predicted usage as it would have done in the past. Ideally, this should be tracked on a daily basis, and any trends away from what was expected should be investigated to determine the cause. Whereas previously it would usually have been a change in expected usage in the cloud it could also be a change in the way the charges are levied.

So, far from capacity planning not being required in the cloud, in order to keep costs in line with expectations capacity planning is even more needed than ever and is more complicated than it ever has been before due to the huge number of different ways to achieve the same result and the complex charging policies used that can be difficult to predict in advance.