Update: May 19, 2015: Google cut prices on compute resources again this week. We've got the latest analysis on Google vs. AWS pricing along with price comparision charts.
Update December 5, 2014: In the 1-Year Partial Upfront RI and the 1-Year All Upfront RI charts, we corrected the plus and minus signs on the percentages for consistency.
Update December 3, 2014: We changed the first chart below to reflect the correct percentage for the No Upfront Savings for 1-Year RIs, which is 20-50 percent rather than 20-43 percent as originally published.
AWS today announced a new Reserved Instance (RI) pricing model in an effort to simplify RI purchases and better match the way enterprise customers want to buy.
The new model eliminates the previous Light, Medium, and Heavy distinctions and instead provides three options that increase the discount as customers pay more upfront.
Below is a summary of the new RI types:
The ranges for savings varies greatly across different regions as well as operating system. With all of the new RI types, the customer is billed for a full 24 hours a day every day of the year, regardless of whether the RI is used or not. This is comparable to the old-style Heavy RIs.
Google vs AWS Pricing Comparison
With the announcement of this new AWS RI pricing model, we’ve updated our AWS vs Google pricing comparisons. The change in AWS RIs does not affect on-demand prices, so Google continues to offer a lower cost since reducing its prices in October 2014.
We compared Google Sustained Use discounts to AWS No Upfront RIs. These options represent the closest to an apples-to-apples comparison, since the Google Sustained Use discount does not require any upfront payment. All prices are based on Linux instances in the US East region. Keep in mind that this is a straight price comparison of similar instances. The specific performance of each instance type may vary.
In this comparison, Google has a price advantage ranging from 11 to 30 percent. However, AWS also offers more memory on the high-memory instances and twice the memory on the high-CPU instances, along with an SSD, that balances Google’s lower price. Depending on workload, the high-CPU c3 series from AWS could provide a better price/performance option.
This also does not take into account potential additional volume discounts for AWS RIs, which start at 5 percent when the annual value of RIs reaches $500K, and tier up to 10 percent when the annual value of RIs reaches $4 million, with higher tiers available on a negotiated basis. It is also important to note that the Google Sustained Use discounts require no commitment at all, while the AWS No Upfront RI is still a commitment to use for a one-year term.
AWS users may also consider using Spot Instances for certain workloads, which can be substantially cheaper, or buying RIs via the RI Marketplace. AWS describes part of the value of RIs as providing a “capacity reservation.” Google, in comparison, states that it is responsible for ensuring that capacity is sufficient.
We next compared Google Sustained Use discounts to AWS Partial Upfront RIs and AWS All Upfront RIs. In these cases, AWS offers a lower price. However, customers will also need to consider the financial tradeoffs of paying part or all of the costs upfront. Excluding this additional “cost of capital,” AWS costs range from 1 to 9 percent lower for the standard and high-memory instance families. However, Google’s cost is lower by up to 14 percent in the high-CPU instance family.
AWS has offered even larger discounts for a three-year term on RIs. However, during that three-year term, customers should consider that there may be significant price cuts in on-demand usage as well as the launch of new, more powerful instance types. These may eat into the projected savings of AWS over the Google Sustained-Use model.
Decoding AWS RIs
While we provide a comparison of Google and AWS cloud pricing, for some AWS users the choice is not about price but rather about which type of AWS instance to buy. Although the RI discount levels calculated by AWS are based on running an instance 100 percent of the time over the term of the RI, it is still possible to realize savings for an instance that doesn’t run 100 percent of the time. It’s important to do a breakeven analysis to determine the amount of time an instance must be used before you break even vs on-demand. After that breakeven point, you start to realize savings for the remainder of the RI term.
The breakeven point for an RI will depend on the discount level. For example, if a No Upfront RI for a particular instance type in a particular region offers a 25 percent discount for a 1-year term, then you will realize some savings as long as you use that RI for at least 75 percent of the year (or at least 9 months). In order to realize the full 25 percent savings, you need to use that RI for the entire year.
We analyzed breakeven points for RIs on m3.mediums. For 1-year RIs, the breakeven point ranged from 222 days (61 percent of the term) for All Upfront RIs to 261 days (72 percent of the term) for No Upfront RIs. Notably for this particular instance, the breakeven point for All Upfront RIs was only four days shorter than Partial Upfront RIs.
We also analyzed breakeven points for 3-year RIs for m3.medium. All Upfront RIs required only 409 days (37 percent of the term) to breakeven while Partial Upfront RIs required 435 days (40 percent of the term) to breakeven. The 3-year RIs offer a significantly faster breakeven as a percentage of the 3-year term. This helps to mitigate the risk from changes in on-demand prices, from the introduction of new higher-power instance types, or from changes in a customer’s cloud usage — all of which could decrease the realized discount over the longer 3-year commitment.
Assuming that your breakeven analysis shows that an RI purchase makes sense, you then need to determine which type of RI to buy. Your company will likely use several additional factors that might include your cost of capital, liquidity requirements, capitalization options, and interest rates to determine whether you are better off paying for an All Upfront RI to secure the largest discount or paying month-by-month for a No Upfront RI.
In some cases, that decision might also depend on other business factors as well. For example, large enterprises might have money budgeted in a particular month, quarter, or year that needs to be used in that time period. Conversely, a startup might want to conserve cash in anticipation of future fundraising.
Considering Future Price Reductions
As you decide whether an RI purchase makes financial sense, remember that purchasing an RI might exclude you from future price reductions that apply only to on-demand instances. For example, if you purchase a 1-year RI that offers a savings of 10 percent over on-demand, and if after half a year the on-demand price goes down by 20 percent, you are now no further ahead than if you had stayed with on-demand pricing. While no one can predict how much cloud prices will decrease over the year, given past history, it’s almost certain that they will decrease.
Other Tips on AWS RIs
It’s also important to remember a few things about how AWS RIs work. You purchase an RI for a particular instance type, availability zone (AZ), and operating system. You can request to modify your RIs to move between AZs (not regions) or to change instance sizes in the same family. You can also modify your RIs between EC2 Classic and EC2 VPC. Requests to modify are processed by AWS as soon as possible based on available capacity.
You can continue to use the old-style RIs over their term, and you can combine old-style RIs with these newer RIs. Your AWS bill will be calculated by applying the lowest cost RI first in order to maximize savings.
Cloud prices from AWS and other public cloud providers will continue to drop, and cloud pricing models will continue to evolve for the foreseeable future. Although AWS has simplified RI prices with this announcement, there are still many factors that companies will need to take into account when deciding how to purchase cloud compute resources. The RightScale 2014 State of the Cloud Report found that most enterprises are implementing multi-cloud strategies. IT teams want to broker multiple public and private cloud services for their organizations and acquire the best cloud services at the lowest possible cost. Savvy cloud brokers can leverage myriad discounting options to optimize, and ultimately reduce, their infrastructure spend.
If you want to analyze the impact of cloud prices on your own cloud spend, get a free trial of RightScale Cloud Analytics to analyze your past cloud usage and create scenarios to forecast future spend on AWS, Azure, Google, and other clouds.
For more on understanding the new AWS RI options and comparing them to old-style RIs plus details on the discounting options from Google and other cloud providers, watch the on-demand webinar, Decoding the New AWS Reserved Instances: What It Means for Cloud Pricing.