Cost Control of the Cloud: On-Demand or Reserved Instances
Cloud IaaS is increasingly a core component of IT operations strategy in the deployment of new technologies to support business goals. One of the most asked questions before starting any cloud deployment is cost control. Being able to extract the maximum value out of your cloud infrastructure, especially those resources/servers that will be online for long expected periods of time would be a great benefit to many companies. The most common way to purchase IaaS instances in public cloud is in an on-demand model. You pay for what you want when you use it at a predefined price for the increments of time that you use it. Additionally, there are two other means of purchasing an IaaS instance in either Microsoft Azure or Amazon AWS/EC2 today which are spot pricing and reserved instances.
- Spot pricing is available in Amazon AWS only and is best used for workloads that are cost sensitive and not time dependent. When the load in a region falls, Amazon will allow for more inexpensive instances to operate due to the available capacity. For computational workloads related to scientific research or investment modeling, this is a great option.
- Reserved instances, which have been available in AWS for some time and just recently announced to be coming to Azure, allowing you to lock in a certain IaaS workload for a term, usually, 1 or 3 years resulting in savings up to 75% depending on a variety of factors. Great uses for these instances are machines that will be online for long periods of time, such as domain controllers, authentication servers (ADFS and Azure AD Connect), or web servers.
AWS and Azure do treat these Reserved Instance contracts differently. With AWS if you want to get out of a Reserved Instance contract prior to its completion you need to use a third-party service to transfer it to another company. Azure is reportedly going to allow for refunds of the Reserved Instance with a minor penalty (Rumour is 1 month of runtime). As the Azure offering details are pre-release, the final terms and conditions are not completed yet, this blog article will be updated once that becomes available.
To illustrate the cost savings below is a cost comparison for Reserved Instance vs. On-Demand. Here is a graph from Cloudyn showing an on-demand instance in comparison to reserved instances:
On Demand and Heavy Utilization Breaking Point
Highlighted is the important area. The blue line is the on-demand cost for a larger instance that grows in use over the course of getting that workload running to the desired performance state, reaching $28,000USD in cost over the course of a year, and continuing at that higher cost into the future. The green line, which represents a reserved instance for the same performance capabilities of the blue line, at full utilization and runtime (i.e. running 744 hours a month or all of the time) shows that in around 6 months the reserved instance pays for itself in this scenario. That is $14,000USD that your organization would not have to pay to receive the same level of service over the course of year 1. Additional savings will be realized in year 2 and beyond.
For the right workloads, those that are predictable and stable, utilizing reserved instances on either Azure or AWS is an excellent strategy to reduce and limit unexpected cloud costs. As discussed there are penalties for dropping out of such a pricing strategy early. In a way, these penalties are similar to long-term telecom agreements. It is a good exercise to make sure this is the right strategy for your company over the course of a year or more.
At Softchoice our group of experts on both Microsoft Azure and Amazon AWS can help you analyze your current cloud environments to find cost savings using Reserved Instances and other methods. We offer a number of services that can help you determine a cost savings strategy for both immediate savings and long term optimization of your costs.