AWS re:Invent 2018 Announcements That Can Save You Money Right Now
The 2018 re:Invent conference is in the books, and the number of new features and services can be hard to keep track of. Here are a few of the announcements that you can start using right away to help reduce your AWS bill.
Hailed as a feature targeting new applications or applications with unpredictable demand, I think that this feature is going to prove extremely cost effective for NonProd workloads and back-office tasks. We run a lot of one-off tasks that make use of DynamoDB as a backing store, but perhaps only see use once per week or even once per month. Previously I would often try to overload a table’s key design to fit more data in the same table, sharing between tasks and saving on cost. Those days are gone! I can have five tables for the price of none now, and I am still only paying for the number of reads and writes I actually consume.
To turn on On-Demand, navigate to the table in the console and select On-Demand from the Capacity tab. For the example below with a 10 read/write table, we’re already on track to save ~$5.81/month.
For those of you using CloudFormation and Terraform, there is newly added support for a BillingMode attribute which can be set to ‘PAY_PER_REQUEST’ in lieu of passing the ProvisionedThroughput element.
As of this writing, the us-east-1 pricing is $1.25 per million writes and $0.25 per million reads.
If you have a lot of data hanging around in S3 (which is most of us), there is a new Storage Class available for consideration. With S3 Intelligent Tiering, AWS will monitor object access patterns over time and transition individual objects back and forth automatically between Standard and Standard-Infrequent Access. If an object goes idle for a long period of time, it will naturally transition over to Standard-IA. If it heats back up, it will be returned to Standard. The Intelligent Tiering will charge you for the object at the Storage Class it currently resides in ($0.023/GB for Standard and $0.0125 for Infrequent Access in us-east-1), plus a small markup for each object under management ($0.0025 per 1,000 objects).
So what’s the catch? Intelligent Tiering is NOT well-suited to short-lived objects — it stipulates that each item will be billed for 30 days minimum. If you already know your data is a good fit for Infrequent Access, that storage class will remain the cheaper option. If you are not sure or have not yet drilled into the exciting world of S3 Lifecycle Policies, Intelligent Tiering could be a great way to save up to half of your S3 storage costs. So how can you get started? Consider adding a Lifecycle Policy to an existing bucket which transitions objects to Intelligent Tiering 30 days after creation.
Increased Performance of General Purpose SSD (gp2) Volumes
PIOPs. When you need ’em, you need ’em, but they sure can be expensive. A good trick for achieving similar performance at around half of the price is to take advantage of large gp2 volumes, which provide three IOPs per GB. Prior to re:Invent, the maximum IOPs for a single gp2 volume capped out at 10,000, leading to a lot of 3,334GB drives floating around. With the new ceiling increase, you can now achieve higher throughput and go all the way up to 16,000 IOPs with a 5,334GB gp2 drive. The comparable IOPs price for 16,000 PIOPs on io1 volumes comes in at $1,040/mo. (not inclusive of the storage costs), versus $533/mo. for the massive gp2 volume.
PIOPs still have higher baseline throughput, and the maximum number of IOPs available has doubled to 32,000 (and 1,000 GB/s) for applications which need extremely high performance. For those that previously needed more than gp2 could provide, consider evaluating whether the newer 16,000 IOP gp2 volumes can meet the need at a lower price.
Amazon announced a new general purpose, low-powered instance family that is optimized for cost. These new instances are priced at about half of a comparable T3 or M5 instance. What’s the catch? These new instances have relatively weak specs and are ARM-based, an instruction set typically reserved for power-sensitive smartphones and raspberry pi. The pitch for these is that they are actually well-suited for horizontally scaling micro-services or scripting micro-tasks which don’t care much about the underlying hardware and aren’t resource intensive. If you have services that fit this description, consider giving the A1 instance family a look.
Cost optimization is an ongoing effort for any organization with a large AWS footprint. While these new offerings won’t single-handedly turn around a runaway AWS bill, they can help alter the trajectory. Do it enough times across enough accounts and the savings can really start to add up. The barrier to access is quite low — if you or your organization are interested in shaving costs, consider taking a look at one or more of the suggestions in this article.
