How to Get the Most out of AWS Instance Size Flexibility
With Amazon’s recent announcement of Instance Size Flexibility (ISF), the savings of a Reserved Instance can now apply to any size of EC2 instance you run inside of a family. Let’s recap how ISF changes RI planning and go through a few best practices that our customers already use.
Last month, we covered how we think this will simplify RI planning and make it easier to achieve savings. Check that out for more details, or read on for some ISF pointers.
What is ISF?
ISF is a new feature of EC2 RIs that lets the savings of an RI apply to any size of instance within a family. You automatically get this benefit when you purchase a regionally scoped Linux RI.
How Does ISF Work?
ISF manages your reservations in a points-based system, where a scale determines the units assigned to each size within a family. When you run an instance, r4.xlarge for example, any ISF RI within the r4 family can apply to the instance and the number of points your RI has, relative to the points for the instance size ran, determines how much or how many of the RI is consumed by that instance hour. To understand what the scale is and how billing works for ISF RIs, read our knowledge base article on ISF.
How Our Customers Are Putting ISF to Work
Our most cloud-savvy customers have already converted their non-capacity constrained RI inventory to be ISF-eligible, by making sure all their Linux RIs are set to be regional scoped.
As Andrew Hatch, DevOps Manager at Seek put it
“Switching our RI fleet over to take advantage of ISF was a no-brainer. Using the streamlined strategy implemented within Cloudability is going to save us a ton of management headache into the future and lead to bigger savings from our RI initiatives.”
ISF Best Practices in Action
The example below illustrates a way of thinking about how ISF can be applied. Each column represents the instances that were run in a given hour. For r4s and c4s, large (L) is the smallest instance size. The concept of targeting a waterline is still completely valid, and if we were to aim for 100% RI utilization in this example you’d purchase 19 large RIs. What we are effectively suggesting here is purchasing ‘points’ at the smallest size within a family. This would mean you’d have only one actual RI purchase which is well matched to your usage. If instance sizes fluctuate, but your overall normalized usage stays the same or increases, your RIs will remain perfectly utilized.
Looking at the RI Planner details page, you can see how we apply ISF to the c4 family.
The Cloudability recommendations engine analyzes your usage and determines the break-even point (as it did before). The histogram we’ve adjusted to represent Normalized Instances, where in this case, we convert all your existing RIs, i.e. 260 xlarge, 18 8xlarge, 8 large, and 2 large (AZ scoped), to represent the number of units relative to a large.
You can also find the breakdown of units for each size in the existing RI table. Once we determine the coverage you will need, we recommend the smallest instance size for the family. In the case of c4s, the smallest size is a large and we recommend you buy an additional 1720 c4.large RIs to cover your usage.
Takeaway: Our ISF Saving Philosophy
We believe it’s best to maximize savings with the least amount of management overhead.
There are a number of ways you could achieve this, but after much internal scenario testing and discussion with our biggest customers we’ve established an effective pattern which we’ve built into the planner. That is purchasing at the smallest instance size possible while maintaining your regular waterline strategy. This allows for maximum coverage with the smallest set of RI parcels.
Note that the parcels themselves will be large, and you may need to request an increase to your RI purchase limit from Amazon to accommodate the purchase of a large quantity of small instances sizes (to find out how to request the increase, click here).
You’ll benefit from a more streamlined RI lifecycle with the purchase of the large quantity, rather than making multiple purchases of a variety of sizes per family. This strategy will also help you better manage your inventory if your usage decreases, with the ability to sell off a much more granular number of RI units to fit the new usage patterns.
“As our cloud grows, managing reserved instance coverage gets complicated,” said Hatch of Seek. “Cloudability gives us RI planning tools that enable us to build the kind of portfolio and strategy that we need to drive as much savings as possible.”
If you’d like to see these strategies at work with your own data, start your free trial of Cloudability today.