At the risk of repeating myself, I’ll say that launching Ocean was a special experience. Our platform grew very organically over years, with our R&D team delivering products as our customers needed them, with little fanfare. Ocean was our first true product launch since Elastigroup, the product that started it all. So when our CEO got up on the stage at re:invent and announced Ocean as the Serverless Kubernetes Engine, it was the culmination of thousands of hours of work from everybody at Spotinst (and many of our customers). So yeah, we all felt all warm and fuzzy inside.
Now that Ocean is out there, helping our customers ship containers without worrying about provisioning or scaling infrastructure, we’re back at work making it better. I’m happy to tell you about the first of many enhancements to Ocean: Predictive Auto Scaling.
The foundational technology for Ocean, is Elastigroup, the automated application scaling service. When it comes to scaling compute for containers, Elastigroup uses a concept called Headroom, which essentially schedules idle infrastructure so that future containers will always have available capacity to run on.
How Does Predictive Auto Scaling Work?
Ocean’s Predictive Auto Scaling capability studies your infrastructure behavior and patterns, and automatically schedules headroom capacity in advance, only when the system anticipates you’ll need it. For example, if your cluster consistently requires the underlying compute to scale up between 4pm and 6pm on weekdays, Ocean will recognize the pattern and adjust the infrastructure as needed, ahead of time, so your containers won’t be even waiting for VMs.
Predictive Auto Scaling is the latest addition to the already powerful Container Auto Scaling that powers Ocean. This technology optimizes container clusters for maximum availability at minimum costs. Beyond spinning up capacity when you need it for new workloads, Ocean ensures your containers are placed in the most efficient way possible. Learn more about it on our docs.
Ocean’s Predictive Auto Scaling capability takes 12 hours to study your underlying compute pattern and apply a model to suit your infrastructure for the next 12 hours. After the initial period, Ocean will continue analyzing your clusters for recurring trends.
Predictive Auto Scaling is available now for all Ocean clusters! There’s nothing to enable or turn on, Ocean will start analyzing your infrastructure automatically, and scale your compute accordingly. You’ll also be able to analyze how Predictive Auto Scaling influenced your infrastructure, with the new dashboard we’ll be releasing soon.