As AWS recently announced, per-second billing for EC2 instances will be available in all AWS regions. This new billing model introduces greater precision to how we are charged for using some popular AWS resources. Since the announcement, Google has been quick to react with its own version. But what does this new pricing model mean for customers and how can they use it?
Where It Applies
The minimum charge for per-second billing is now one minute rather than one hour, which will likely result in us paying less for compute instances going forward.
AWS per-second billing will apply to Linux On-Demand, Reserved, and Spot EC2 instances. Per second billing for Reserved instances is, in my opinion, a major benefit and makes these instance types even more attractive for predictable workloads. Per-second billing will also apply to AWS Batch and Amazon Elastic Map Reduce. This should also reduce the costs of running AWS Batch since few jobs last a whole hour.
Per-second billing does not include Microsoft Windows instances or to all Linux distributions, so some Linux AMIs may still have an hourly charge. Always check the simple monthly calculator for current unit prices.
When researching instances, one resource that I like to use is this EC2 Instances website. While it is an unofficial EC2 instance guide (always cross check with the official AWS site before making any purchase decisions), the site has also included per-second pricing. It’s a great time saver.
Benefits of Per-Second Billing
Google Cloud Platform has also announced per-second billing for compute instances, which is great for us as customers. Google App Engine, Compute Engine, Container Engine, Cloud Dataproc, and virtual machines are now billed per second. Microsoft Azure provides per-second billing for Azure container instances.
So, what are the benefits for us? In my opinion, this level of granularity will reduce costs for large, spiky workloads where resources are being scaled up and down to meet demand. Think retail sites, business apps, big data, analysis, or queue processing.
One-off workloads stand to benefit as well. For scenarios where we want to spin up large instances for specific tasks such as data parsing/crunching, we will be closer to paying for the actual compute time we use. For steady/predictable workloads, per-second billing is probably least likely to reduce costs. In the future, I would like to see per-second billing applied for Elastic Block Store snapshots.
It’s always been a slight bugbear of mine that I had to pay for a whole billing hour even if I used an instance for just 15 minutes! The biggest benefit of per-second billing is that it brings us closer to being billed ONLY for the time we actually use AWS and Google Cloud Platform resources! That is truly a benefit for everyone.
The new pricing model will be effective on October 2. Check the official AWS announcement for more details.