This is the third part of our series on rightsizing.
In the second part, we discussed the complexities around rightsizing and why an automated tool is usually the right answer. Now, we will dive deep into burstable series offerings, and try to understand the cases in which burstable can be the right option.
From our experience dealing with rightsizing for customers, it is a general feeling that clients are not comfortable using burstable series for workloads. We try to change that perception by looking into their historical CPU/ Memory/ Network metrics and then displaying visually how a load can be an ideal candidate for the Burstable series.
For burstable instances, we look at your CPU utilization to determine if the burstable instance can cover the load at the maximum peak and the burstable instance base CPU utilization is greater than the average CPU Utilization.
Average CPU Utilization <= Burstable Base Utilization
Maximum CPU Utilization <= Burstable Max CPU capacity
If yes, then we mimic the workload on the old instance as a burstable instance to determine whether the peak CPU utilization can be covered with credits earned when the CPU is running at low utilization. The below table shows the base CPU, max CPU, and credits earned for some of the burstable instances for both AWS and Azure:
To understand the usage of credits this image might be helpful:
There is another important feature in burstable instances, which makes customers much more comfortable using burstable series in AWS.
T3 Unlimited Mode
If the instance runs out of CPU Credit Balance, surplus credits consumed above the maximum are charged at a rate of $0.05 per vCPU hour — AWS will allow the instance to burst at $0.05 per CPU hour, meaning that a T3 Unlimited instance, such as a t3.nano could use 100% of CPU for as long as needed, at an additional cost of $0.05 per hour.
Breakeven Point for a burstable instance with unlimited credentials represents at what percentage of average CPU utilization will the price of the old instance be equivalent to the new burstable instance.
Let’s understand this with this visualization from AWS’s whitepaper on pricing:
In this example, we see that the old instance configuration was an m5.large and the workload is spiky with the average value staying below 30%. Hence, replacing it with a t3.large can be an option that provides a breakeven point of 42.5% CPU Utilization.
At Cloudwiry, we can help you rightsize the instance with the click of a button at the time of your choice with our Rightsizing Automator.
With this, we come to an end in our series on Rightsizing. Want to read more about cost-saving on AWS? Read our APN blog on using Convertible Reserved Instances to maximize savings.