Blogger social buttons

Tuesday, 5 February 2013

AWS Reserved Instances vs On-Demand: Breakeven point

When deciding whether to use AWS Reserved Instances, 
evaluate the breakeven point in terms of months per year

AWS Reserved Instances are a purchase option offered by Amazon for their EC2 instances and their RDS databases. Instead of paying the normal hourly on-demand cost, you can choose to pay a lump sum upfront and get a reduced hourly rate. The upfront cost and the hourly rate depend on a number of different things, which you have to commit to at the time of purchase:

  • Instance Size
  • Operating System
  • Region (e.g. US-East) *
  • Utilization level
  • Purchase duration
* Note that when purchasing Reserved Instances, you actually have to choose which Availability Zone within that region you would like to commit to.

Reserved Instance Utilization Options
There are two terms that you can choose when buying Reserved Instances: 1-Year and 3-Year. Each of these terms comes with three different utilization options:

Light Utilization: This has the lowest upfront cost, but the least reduction in hourly price. You will only pay for the hours which you have instances running.
Medium Utilization: This costs a little more upfront than Light Utilization, however you get a more discounted hourly rate. Again, you will only pay for the hours which you have instances running.
Heavy Utilization: This has the highest upfront purchase cost and gives the lowest hourly cost. With this option however, it does not matter how long you run your instances for, you will have to pay as if they were running 24x7.

Calculating Breakeven Points for Reserved Instances
In order to assess the breakeven points for each of the utilization levels, I have run scenarios through PlanForCloud for an m1.Xlarge (Standard Extra Large) on US-East running Linux for a 1-year term.  You will see similar breakeven points for other instance types (e.g. m1.medium or m3.2xlarge).

Here are the results, the breakeven point from On-Demand to Light, Medium and Heavy utilizations are:

AWS Reserved Instances breakeven points

This means that if you are planning on running the instance for:
  • Below 3.5 months in a year, just use On-Demand
  • Between 3.5 months in a year to 5.5 months, buy Light Reserved Instances
  • Between 5.5 months in a year to 7 months, buy Medium Reserved Instances
  • Anything more than 7 months in a year, buy Heavy Reserved Instances

[5 March 2013 Update] AWS have updated their EC2 Reserved Instance prices and therefore the breakeven figures on this blog are no longer accurate. However, price in the PlanForCloud tool have been updated, therefore you can use the tool to calculate the breakeven points for your specific use case.

There are two additional questions to consider:
1. Do you expect to maintain the breakeven usage levels in the above chart over the entire term of the Reserved Instance?  If the usage level of your instances may decline over the life of the Reserved Instance, you may want to consider purchasing a lower utilization level for your reserved instance.

2. Do you want to pay the upfront cost associated with the optimal number of Reserved Instances? Smaller, growing companies may want to preserve cash by avoiding any upfront payments.  Even larger companies may be constrained by capital and prefer a pure pay-as-you-go approach.  In addition, for larger deployments that require a significant upfront payment, there is a cost to pay upfront that could shift the breakeven point.

An approach we often see from users buying reserved instances for the first time is the following; buy sufficient reserved capacity to meet the average sustained usage of one tier of an application. For example, in scalable array web servers, identify the minimum number of servers necessary to “keep the light on” and reserve that many.

To help you answer these questions and much more, PlanForCloud is a free tool that helps you model your cloud infrastructure requirements and get a detailed cost report to see which purchase options make sense for you - Log In as a guest and try PlanForCloudl

A final note: if you are designing a Disaster Recovery procedure, reserved instances will guarantee you availability in your chosen availability zone, read our blog post about The Cost of Keeping your Application Running During Cloud Outages

-- Hassan Hosseini
Product Manager at PlanForCloud

1 comment:

  1. @hassan How you calculated breakeven point for On-Demand to Light, Medium and Heavy utilizations?