Impact of per-second AWS billing blunted by changing uses

AWS per-second billing is a welcome change, but one likely tailored to lure new enterprises, rather than assist users that have optimized their environments or gone up the stack.

AWS billing has finally caught up to other vendors' pricing with more granularity, but the shift may be less important...

these days, with savvier customers and the emergence of serverless computing.

Amazon forced its way into the IT market more than a decade ago with per-hour billing. Then, Google and others offered even greater refinement through per-minute pricing. Now, starting Oct. 2, AWS will leapfrog those cost structures and charge Elastic Compute Cloud (EC2) instances and Elastic Block Store (EBS) volumes on a per-second basis.

But more granular pricing structures have less impact for longtime users that have already optimized their usage and as AWS and others push advanced services that abstract the underlying machines.

Part of the allure of the public cloud is the ability to spin instances up and down as needed, but many users that run production workloads on AWS do so with websites that must remain available at all times.

Realty Data Co. kept its traditional client-server approach for its web server when it moved to AWS, but shrank its overall footprint to improve efficiency with AWS VMs. The financial services company, based in Naperville, Ill., runs most of its servers 24/7, so it doesn't particularly matter how the day's usage is sliced up by AWS, as long as the bottom-line price is the same.

"The only way I could see it helping is if I tried to get better utilization by scaling up and down based on the end of the month, but I've been able to scale down our system because the instances are pretty powerful," said Craig Loop, director of technology at Realty Data.

Customers like Realty Data have tuned their systems based on selecting the best machine type for specific workloads, but the per-second AWS billing eliminates the need for that level of sophistication, according to Amazon.

ACI Information Group, an aggregator of social media and editorial blogs in Ipswich, Mass., runs its older workloads on Amazon EC2 and builds its newer applications on AWS Lambda. The per-second billing could push more enterprises to use Lambda, because it can link to more efficient use of EC2, but ACI hasn't built its environment to connect the two services.

"It isn't insignificant, but it's not as important as it would have been before we locked in our architectures," said Chris Moyer, vice president of technology at ACI Information Group and a TechTarget contributor.

ACI also runs many of its EC2 instances constantly, but there could be marginal savings with how Auto Scaling responds to VM failures, he added.

"If something went wrong and a new instance didn't spin up in time for Auto Scaling to identify it, it could end up spinning up 30 instances in an hour and charge for 30 hours, even though it's only up for five minutes," Moyer said.

AWS billing change still significant

Regardless of broader trends or the lack of novelty around this move, it is certainly a major change for AWS. Amazon has cut costs for specific services many times over the years, and it has added discounts for deferred use, upfront payments and secondary market purchases. But this is the first change to the underlying cost structure for VMs since EC2 was launched in 2006.

And while this move is less important than it may have been in the past due to the advent of serverless computing, per-second AWS billing will still translate to savings for a large cohort of users. This will be particularly good news to a user who runs a batch job that terminates in less than an hour or a test-and-development scenario where the user is charged for the full hour every time they start and stop an instance, said Dave Bartoletti, an analyst with Forrester Research.

"Enterprise developers are just beginning to tap into the value of functional pipeline programming, microservices, etc., but there are millions of workloads that will run in operating systems on infrastructure for years," he said.

Amazon claims the real benefit won't come from price reductions, but rather from more experimentation with EC2 and new ways to utilize the second-oldest service on AWS. It also could advance adoption, or at least help control costs, for some of those higher-level services, industry observers said. For example, Lambda, which is priced per millisecond, can be linked to EC2, and many of the AWS machine learning tools are tied to EC2, as well.

This will ripple through the way DevOps teams think about orchestration and automation ... and push the whole industry to a much more granular way of thinking about compute.
Greg Arnettefounder and CTO, Sonian

"This will ripple through the way DevOps teams think about orchestration and automation of spinning up servers and taking them down and offline, and push the whole industry to a much more granular way of thinking about compute," said Greg Arnette, founder and CTO at Sonian, an archival storage and analytics company in Waltham, Mass., that built its service on AWS.

Some users that want to optimize their costs will need time to adjust, as will third-party management vendors and their users, as those tools are tailored to the hourly cost structure, Arnette said.

"When we knew we bought an hour, we didn't have to be so rigorous. But with the switch to per-second, we'll have to do some modeling within our cost management tools to see where we can eke out more savings and if we should be more aggressive on launching and terminating compute instances," Arnette said.

Those who use the Spot Instance market, sustained usage discounts and Reserved Instances also need to adjust. AWS will bill for Reserved and Spot Instances by the second, but still list them on a per-hour basis. Users can't get a refund if, for example, they purchased a single hour and used only 40 minutes, but can apply those unused 20 minutes to other instances during the same hour.

The per-region fee for Dedicated Instances remains billed by the hour, as are EBS Snapshots and products in the AWS Marketplace. The new AWS billing model is also limited to Linux machines, which means it's not available for VMs that run Microsoft Windows or Linux distributions with separate hourly charges.

And while some longtime AWS observers weren't blown away by the per-second billing change, they did think it could portend bigger changes to come. Given Amazon's track record with new products, this could be a piece of a new service or services that will be rolled out heading into AWS' annual user conference a little over two months from now.

Trevor Jones is a senior news writer with SearchCloudComputing and SearchAWS. Contact him at [email protected].

Next Steps

Get control of your AWS budget

Turn to these native, DIY tools to optimize AWS costs

Adhere to these AWS cost optimization principles

Dig Deeper on AWS pricing, cost and ROI