One particularly important session during re:Invent was Billing and Cost Management (re:Invent registration required). In this session Keith Jarrett, the head of cost management product marketing and business intelligence at AWS, explains the fundamentals of cost management and introduces some new services that AWS has rolled out recently.
Keith lays out the major questions you have to think about when dealing with cost management including:
“How do we have the necessary visibility into our cost and usage to meet and make sure that we understand our reporting and analytic needs?"
"How do we understand and identify opportunities to lower or optimize our cloud spend?"
"How do we adopt budgeting and planning processes that could be more agile or predictive?"
"How do we manage all this at scale?"
"How do we establish the necessary guardrails and make sure we’re rapidly innovating, but avoiding unnecessary cost surprises along the way?”
These are great questions and the new services introduced in this talk, which include Budget Actions and Anomaly Detection, are genuinely exciting. At the same time, we get more information on the what of new services than how you’d use them to manage costs in the real world. Since the new services are the real meat of the talk, I’ll focus on giving you the key details of those, while filling in some of the gaps on how you can really use them for cost control.
Budget actions are a new feature that let you set up automated actions that trigger if you go over budget. This is a cool feature, but there are some challenges with putting it into practice.
You’ve got a few options for how to react when it detects you’ve gone over budget. You can cut off the user or group from spinning up new instances, send out an alert, or even shut down the instance automatically. “This is a super powerful capability so we actually recommend using an approval workflow here.”
Have you ever seen a movie where someone who’s being held hostage is talking on the phone, and they can’t say they’re being held hostage, but they try to signal it using some kind of code? That’s what I feel like Keith is doing here. By “super powerful,” I’m pretty sure he means, “really dangerous.”
Indeed, would it really be a good idea to automatically shut down instances based on budget metrics? Clearly you don’t want to shut down essential parts of your architecture like web servers or production databases. You might want to shut down something like a development sandbox, but even in that case, if the instance is being used for valuable work, it might be preferable to go over budget than to blow up someone’s work midstream.
If you want to save costs by shutting down instances, we think it makes more sense to start with unused or underutilized instances. That way you can save money without shutting down processes that are providing real value. That’s why we automatically send you alerts when we detect unused or underutilized instances.
The next new feature is anomaly detection and Keith goes into some of the technical details on this one. Their system is built on top of forecasting and uses the residuals (the difference from the expected value) to detect anomalies. That means you could have a situation where your cost goes up, but it’s not an anomaly, because it was an expected increase; or your cost stays constant, but based on your history it really should have gone down during this period, so it is an anomaly.
That’s pretty cool! The only catch is, it means the anomaly detection is only as good as the forecasting. As anyone who’s tried it knows, predicting the future is hard. To see if your usage on a Wednesday this December is out of whack, they might compare it to last December. But were you using AWS in the same way last December? If not, will the model pick up on that?
Maybe they’ve figured out ways to address issues like these with their modeling, but I’d want to see more data before relying on it. In the meantime, we offer our customers anomaly alerts out of the box, and we’ve got some exciting ideas in the works to alert you to unfolding anomalies before you get hit with a higher bill.
How Reserved.ai can fill in the current gaps
The new features introduced in this talk, especially Budget Actions and Anomaly Detection, are exciting and have a lot of potential. At the same time, it’s not easy to understand how to use them effectively, especially when combined with existing features like Cost Explorer, Cost and Usage Reports, Savings Plans, Reserved Instances, Compute Optimizer, and the list goes on.
That’s why we do the heavy lifting for you by searching across all the available savings opportunities on AWS to find the best deal for your infrastructure. If you want to lower your AWS bill but don’t want to spend weeks or months getting into the weeds of AWS services, get in touch with us!