Many companies using Amazon’s AWS cloud services are not capitalizing on the unique pricing model that AWS provides them with.The pricing model, which is specifically geared to enterprises with a major web presence, allows them to dynamically deploy resources in accordance with current traffic levels. Despite that, many companies are not using this model, and are missing out on an opportunity that could save thousands of dollars per year.
Though the Amazon model is not new, many companies have not chosen to adopt it. The reason behind this lack of uptake may be because this practice falls within a gray area between operations and pure business decisions. In addition, the pricing model may appear daunting and difficult to manage. But after taking that initial plunge into the world of AWS cloud services, you’ll realize that it is quite simple to understand.
The AWS pricing model is divided into 3 categories: On-Demand, Reserved Instance and Spot Instance. The most popular category is On-Demand. On-Demand means, according to Amazon’s definition, that you “pay only for what you use” — or put it another way, it is simply an hourly use fee. You can find the pricing table for the instances you are looking for here. The listed prices are based on the region in which your instance is running.
On-Demand vs. Reserved Instance
For example, in the U.S. East (N. Virginia) region, a c4.4xlarge instance for Compute Optimized use — a 16-core vCPU with 62 ECUs (Elastic Compute Units), 30 GB of memory, and an EBS (Elastic Block Store) will cost $0.838 per hour, or $7340 per year for Linux/Unix usage. If you use 150 instances, the price could go up to $1.1 million a year.
What if there is a great demand for that specific instance and Amazon can’t attend to your request? No doubt you will be relieved to know that there is another way; you can rely on the AWS cloud service. If you know in advance that you will need that instance for the next 1–3 years, why not secure its usage and earn a significant discount? This is the great benefit offered by the Reserved Instance category.
By using the Reserved Instance (R.I.) option, you can preserve the instance you want for a period of one or 3 years, and pay less per hour of usage. According to Amazon, you can save up to 75% off the original On-Demand instance pricing. This category is divided into 2 subcategories: Standard Reserved Instances and Scheduled Reserved Instances. In this post, we will focus only on the Standard Reserved Instances.
Different pricing on Reserved Instance model
Standard Reserved Instances means that the instance you choose to purchase will be “always-on”, as Amazon defines it, meaning your instance will work 24×7. There are 3 payment options for this subcategory: All Upfront, Partial Upfront and No Upfront.
For All Upfront, you pay a fixed price in advance. Using our c4.4xlargeexample with a one-year reservation, you will pay $4,607 for the whole year (or $0.5259 per hour). Compared to the On-demand price (which was $0.838 per hour), you will receive a 37% discount.
With the Partial Upfront option, you will make a low payment in advance and then pay a monthly discounted fee. In the c4.4xlarge case, it will be $2,361 in advance, and a $195 monthly fee ($0.27 per hour for the monthly fee). This means that the effective hourly pay will be $0.5365 per hour, providing you with a discount of 36%.
The No Upfront option does not require any in-advance payment; rather it provides an hourly discounted rate. If we continue with the figures given in our example above, you’ll pay a monthly fee of $453 (or $0.621 per hour), resulting in a 26% discount.
You can see it quite clearly in the following table taken from AWS website (for the c4.4xlarge example in U.S. East region):
In addition, you can receive an extra automatic discount on a volume basis. If you purchase a Reserved Instance in the range of 0.5–4 million dollars, you will automatically receive an additional 5% discount. In the range of 4–10 million dollars, you will automatically receive an additional 10% discount. For a purchase volume of over 10 million dollars, you should contact Amazon directly in order to arrange a larger additional discount.
Even though Reserved Instances can provide substantial savings, it is not always recommended to buy all your instances according to that model alone. If conditions change — and suddenly you need only 50% of the instances planned, you will be forced to the pay the amount you committed to in your one or three year contract. In the On-Demand model, you can drop those instances whenever you want and stop paying for them. That is why the optimized model will involve risk diversification; a combination of R.I. and On-Demand.
The following chart will demonstrate the risk diversification according to random instance usage I choose. You can see that the combination of On-Demand and R.I. is the most financially worthwhile.
And there’s another way…Spot Instance
Learning about the On-Demand and Reserved Instance pricing models can save your business a lot of money — but there’s one more option you can use to optimize your AWS pricing model, called Spot Instance (S.I.). This option is best suited for short-term usage since it is based on available instances and the demand for them. Amazon enables you to bid for its unused instances. The prices are really low, but could change in an instant according to the levels of supply and demand. For example, at the time this post was written, the spot price for a c4.4xlarge instance in the U.S. East (N. Virginia) was $0.1616 per hour — 80% off the On-Demand price.
To use Spot Instances, you should place a request stating how many instances you would like to run and the maximum price you are willing to pay for them, based on Spot Price history. If the price you bid exceeds the current Spot price, the request is accepted, and your new instances will run until you choose to terminate them, or until the Spot Price rises above the maximum price that you bid.