Deciding an Approach to the Cloud: AWS Reserved Instances

2 Flares Twitter 0 Facebook 0 Google+ 2 LinkedIn 0 Email -- 2 Flares ×

When starting to use the public cloud, for most businesses AWS is almost always a first choice. It’s very easy to deploy, it has a rich API and plethora of services, and the intuitive on-demand pricing seems very attractive. To the average user it appears that with the simplest on-demand price you pay for what you use by hour.

However, what is soon discovered after some use is that this seemingly attractive price is in reality quite hefty. AWS usage rates often quickly exceed planning parameters and IT departments are soon hit with a huge, unexpected bill.

Amazon offers several ways to reduce this price. A very attractive pricing model is called “Spot Instances”, but it does not fit every purpose – especially when high instance availability is required. Exploring the “Spot Instances” model and understanding when it is best used is worth a separate article, so stay tuned.

Instead, in this post we’ll focus on the “Reserved Instances” pricing model. It offers great value and suits most common cloud deployments, but many businesses are not aware of its benefits and therefore do not leverage it.

How does it work?

With the reservations approach, you make an initial one-time payment, after which hourly cost is reduced significantly. Instances can be reserved for two terms: 1 year or 3 years. The only difference between the two options is the up-front payment.

Reserved instances can be purchased using the AWS Management Console, Elasticfox or API tools. After having purchased the reserved instance/s, prices for the first applicable instances are calculated at ongoing reserved rates.

So many options, so much confusion

A few months ago Amazon further complicated its reservation model by introducing three new reserved instances types: Light Utilization, Medium Utilization and Heavy Utilization. The key differences between each type are the up-front fees and hourly rates.

According to Amazon, when looking at a one year term of engagement, the Light option starts paying off when the instance is up >32% of time, Medium after 48% and Heavy after 61%.

UPDATE: Following the March 6 AWS price reduction, Amazon has changed the above mentioned breakeven utilization points to: Light Utilization starts paying off when the instance is up >35% of time for a 1-year reservation term, Medium after 69% and Heavy after 85%.

The real calculation of usage is even a little more intriguing. The diagram below compares the cloud cost over time of all reservation options vs. on-demand for m1.large Linux instance in us-east-1 region for the term of 1 year. (Full spreadsheet may be found here):
ri-chart-m1-large-linux-us-east-1

Diagram’s key insights:

Cloud instances usage term Purchasing option offering the lowest price
Up to ~2.5 months/year (<23% of time) On-Demand Instances
2.5 to 8.5 months/year (23%-70% of time) Reserved Instances – Light Utilization
8.5 to 10 months/year (70%-83%) Reserved Instances – Medium Utilization
More than 10 months/year (>83%) Reserved Instances – Heavy Utilization
  • Significant savings (>52% annually) can be achieved by using the Heavy Utilization RIs option on an instance that is constantly up for a period of one year.
  • Medium Utilization RIs are almost never the most cost-efficient option.
    • This might sound odd, especially since Amazon is saying that economical advantage is achieved for Medium Utilization instances that are up for more than 48% of a 1-year term. Amazon’s claim is correct when comparing it to on-demand pricing, but when comparing to other RIs types, we found that Light Utilization RIs are a better choice even in this case.
      UPDATE: After having changed their prices, Amazon numbers on the Light Utilization instances are correct. But for a 1-year term period I doubt that you could really tell whether your instance is up 70% – 83% of time. This period is probably shorter than the estimation error.

If you’re interested, you can also find in our spreadsheet a complete analysis of the 3 years usage term.

Reservations Caveats

While the reservation model does cut cloud cost significantly, there are three main caveats to watch for when using the model:

  1. Reservations are for a specific availability zone (not region!): A very common mistake is to reserve the instance in one zone, and to launch it in another.
  2. Reservations bind you to specific instance types: If you reserve one type, and then find out that actually you need a different type, you have wasted the reservation up-front payment.
  3. Heavy-usage reservations are actually a subscription: instance hours are charged even when the instance is not running.

Until recently, Amazon did not allow you to change your reservations in any way after the initial purchase, but this policy is not as strict in recent months. We are seeing customers who are allowed to cancel reservations that were made incorrectly and even get the refund for it.

We also see some (larger) customers getting a special, non-published reservation offering: these customers make a somewhat higher one-time payment and do not pay for the instance hours. That essentially is an equivalent of a 1- year hosting.

Using Cloudyn to leverage cloud economics through the reservation model

Cloudyn’s system was created to help you find the most cost- efficient  pricing option for your deployment needs and scenarios. It  forecasts your expected instance usage based on past use, and suggests the most suitable instance type (offers you to downscale your instance if you do not use its full capacity) and pricing option.

Even though Amazon offers a 3-year reservation model, Cloudyn does not recommend this model. The cloud landscape is changing every day and we believe that in most cases no one is able to estimate the correct IT conditions for such a long period.

Reservations can be real money-savers!

Many organizations do not reserve, expecting the computing needs/conditions to change in the near future, but practice shows that the instances tend to run for much longer than anticipated. For any business that is serious about using the cloud, the reservation model will enable a greater level of control and cost predictability.

Using the Cloudyn system you can gain the insight and recommendations needed for making the right choice, but when unsure – at least use the Light reservation option.

Understanding the various options is critical to keeping your business agile and the costs of cloud computing under control.