256 karma · joined March 12, 2018
kaveh@usage.ai
Looking forward to answering your questions!
The product my team works on, www.usage.ai, automates the process of finding AWS savings using ML (through reserved instances, savings plans, spots, and rightsizing). Every recommendation is shown on a sleek webpage (we spent a lot of resources on UX).
We haven't fully explored the ML use case, but I'd love to figure out how we can help you drive down the costs associated with your GPU models. Would you have 15 minutes this week for a discussion?
If you're interested, you can reach me at kaveh@usage.ai
2. Rightsizing and spot fleet management keep our users coming back - but our customers get the most value the first month. Almost all of our customers have renewed their subscriptions after the first month so far.
3. I don't know how well the computer optimizer model performs, but we haven't had a customer take a recommendation and regret it later on. For our enterprise plan, we fine-tune our models to a company's historical usage (rather than throw a standardized model at all of our users).
Happy to share a screencast! Feel free to email me if you'd like.
Additionally, our growth and enterprise plans all have guaranteed savings. 3x and 5x, respectively.
For instance, if you chose the growth plan ($5k/month) and saved less than $180k after 12 months, we'd give you a 100% refund.
Similarly, if you chose the enterprise plan ($10k/month) and saved less than $600k after 12 months, we'd give you a 100% refund.
I'd say the biggest difference is that we handle everything (so that engineers can focus on actual product dev) and also offer white-glove service to all of our customers. For midsized and up, we offer 24/7/365 access to an engineer for any questions or problems with the Usage dashboard.
I would be interested in hearing about your experience with cost optimization. My email is kaveh@usage.ai if you'd like to reach out.
1. Annual recommendations (Reserved Instances, Savings Plans) Sure, if a company only wants these recommendations - they can bring on a contractor once per year to do these changes. This is the one-off fee type contractor the poster is talking about
2. Continuous recommendations (Rightsizing, Spot Management) These recommendations are typically not something you could do once and be done. Rightsizing (which includes zombie instance detection) can happen through time. Spot Fleet management is also a continuous play - spot prices fluctuate and Usage's strategy works hard to maximize savings while minimizing downtime.
You might ask, how does this compare to hiring a contractor full time?
If you're a midsized company, Usage charges $5k/month (or $60k/year). A fulltime contractor in SV would likely be at least $100k/year and that doesn't factor in the number of billable hours (through interviewing) it would take to find someone who knows 50%+ of all of AWS' features.
Good question!
With Usage, every enterprise customer gets their own fine-tuned model. AWS compute optimizer throws the same standardized model to all of their users. I can't compare our models to the ones AWS compute optimizer uses - but I can tell you that we've never had a complaint about a customer taking a recommendation and regretting it later on.
Additionally, we have humans look over every recommendation to ensure quality. Typically we have an understanding of a customer's cloud infrastructure before they sign up for Usage - and the human in the loop uses that information to grade recommendations produced by Usage's models.
We're more than just RIs. We also offer Savings Plans, Rightsizing, and Spot Fleet management.
Our permissions include CPU, network, and memory logs. We can scale permissions back on a customer-by-customer basis, though (as we have in the past).
Every recommendation is first produced by a model. Then for a recommendation, we add a human to the cycle that understands the client's infrastructure and adjusts the confidence score accordingly. The savings are then shown on a user's dashboard - alongside an analysis report.
When we onboard a new customer, we spend time understanding their infrastructure plans. Then we keep that information for our human-in-the-loop.
I faced a similar issue at the company I work for (where a PM had activated an instance for months and racked up a $9k bill). I've been spending my coronavirus time hacking together a service that hopefully makes billing on AWS more clear. It does cost optimization and anomaly detection right now, and I plan on adding more features as I start to see more use cases.
Check it out if you'd like, and let me know what you think!
www.usage.ai
Great point! It is actually linked to current expenditure. When you first register for an account and connect your AWS, we compute your potential savings and then take 10% of that number spread over a year. So you're not given an unknown number! I'll make that more clear in the FAQ.