Amazon On-Demand Vs. Reserved Instances: Cost Comparison

With rapid price cuts rampant these days, long-term contracts that freeze prices for cloud services over a long period might look like a bad bet. Doubters ask, why not just go with on-demand services and ride that hourly price decline down its slippery slope?

Several knowledgeable sources have raised that question to support the idea that locking yourself into a long-term contract -- such as Amazon Web Services' three-year contract for Reserved Instances -- is a losing proposition. For one example, see GigaOm's article, "Where's My Cloud Price Cut?"

But cloud cost and analytics firm Cloudability has plotted the cost of Amazon Reserved Instances through a series of steep price cuts and found that Amazon has consistently kept its long-term RI servers priced lower than on-demand, by-the-hour instances, even during periods of rapid price cuts.

"An AWS user might hesitate to commit to a three-year Reserved Instance. In fact, this is one of the most common concerns that our customers express," said Mat Ellis, founder and CEO of Cloudability, in an interview. Cloudability customers spend $50 million to $100 million a month on Amazon services, and Cloudability used its billing data dating back to December 2005, when AWS's EC2 was first available in beta form, to plot what one-year and three-year Reserved Instances cost versus their on-demand counterparts.

[Want to learn more about Amazon's latest price cut? See Amazon Strikes Back With Lower Cost Instances.]

Reserved Instances carry a much lower per-hour rate than on-demand virtual servers, but they also require a down payment upfront on the capacity being reserved. That makes it complicated to compare the prices of on-demand and RIs. And there's a third factor that must be plugged in: Amazon's willingness to grant volume discounts, which happened in at least one case. Such multiple price moves lead some industry watchers to suspect that Amazon is cutting prices less than it indicates. The smart thing to do, the critics say, is to simply ride the price reduction wave as long as possible with on-demand instances.

"As a strategy for managing your costs, that's a losing one," asserts Ellis, who heads what is now a 40-employee company in Portland, Ore.

Cloudability tracked the price of an m1.large server, an instance type that's been widely used since 2006, to compare its price during different time periods against that of the same m1.large server as a Reserved Instance. The firm concluded that on-demand prices have consistently remained higher than the Reserved Instance unit price at any given time, even when the upfront cost of the reservation is amortized across the reservation period. (Keep in mind that it's difficult to duplicate this exercise now because Amazon replaced m1s with m3s at the start of 2014. M1.large instances are still in use, however, and Cloudability has the ongoing pricing data on them.) Here are some details of the study:

In February 2013, Amazon cut prices on Linux m1.large on-demand instances by 20%, from $0.26 per hour to $0.208 per hour. A customer who opted to replace the regular hourly rate with the Reserved Instance the day before the price cut would have paid $1,201 for its use over the course of a year. A customer opting for on-demand at the 20% reduced price would have paid $1,458, according to the Cloudability report.

Even in light of the (20%) price drop, the Cloudability analysts determined, the old Reserved Instance would still save $257 over on-demand. The takeaway? Running on-demand all the time in anticipation of price cuts has historically not yielded the greatest possible savings.

In fact, the price on m1.large Reserved Instances decreased at the same time, but by a smaller amount: from $0.064 per hour to $0.056 per hour, or 12.5%. Cloudability's analysis doesn't use this reduced pricing to test whether there could be an advantage to on-demand prices for customers who committed to Reserved Instances just before a price cut.

Cost comparisons between on-demand and Reserved Instances are complicated even when no price cuts are involved, and they're even more so when on-demand prices are dropping rapidly. Amazon has insisted Reserved Instances are the best deal after each price cut, and its Trusted Advisor cost calculator reflects that. The trouble is, Amazon is plugging the numbers into Trusted Advisor, and its recommendations don't necessarily represent "best practices" for every customer.

"Customers are wary of a vendor's recommendations," Ellis acknowledged. But Amazon wants Reserved Instances to both appear as and be the lower-priced option -- it helps Amazon's capacity planning to know how much its customers are planning to use EC2 and other services.

When it comes to building out their own data centers for the next three years, IT managers don't hesitate to overprovision because it takes so much lead time to purchase and install equipment. When it comes to using the cloud over the same timespan, however, they hate to be locked into overprovisioning.

"As big companies buy more cloud services, they will get better at it. They know if they don't get good at it, they'll be beaten out by the companies that are," Ellis said. At the same time, he added, "Sourcing this stuff at the right price" is a skill that's just starting to get established, and opinions among the CIO, CTO, and IT staff members vary widely. Cloudability conducted its analysis to help determine recommendations for its largest clients, Ellis said.

In its ninth year, Interop New York (Sept. 29 to Oct. 3) is the premier event for the Northeast IT market. Strongly represented vertical industries include financial services, government, and education. Join more than 5,000 attendees to learn about IT leadership, cloud, collaboration, infrastructure, mobility, risk management and security, and SDN, as well as explore 125 exhibitors' offerings. Register with Discount Code MPIWK to save $200 off Total Access & Conference Passes.