Amazon has turned machine learning loose on its EC2 service for those customers who find working out likely peaks and troughs in demand themselves is just too much like the old days.
The existing Auto Scaling feature in AWS EC2 allows customers to use schedules and/or rules to automatically ramp up capacity when they need it.
However, this clearly doesn’t make life easy enough for users, so AWS is now offering users the option of switching on Predictive Scaling, which uses “well trained” machine learning algorithms to predict demand.
AWS said it “predicts future traffic based on daily and weekly trends, including regularly-occurring spikes, and provisions the right number of EC2 instances in advance of anticipated changes.”
Over time, it continued, this should remove the need for admins to manually tweak auto scaling at all.
AWS chief evangelist Jeff Barr gave more detail in a blog post, saying the model would start making predictions based on one day’s worth of data, and will then schedule “minimum capacity”. This is re-evaluated every 24 hours to create a forecast for the next 48 hours.
The predictions will be based on customers’ own usage, “further informed by billions of data points drawn from our own observations.”
Predictive scaling works across three metrics: CPU utilization; total network in; and total network out.
Barr added that the feature can be used in conjunction with dynamic scaling, and that a buffer time can be set so that instances are warmed up and ready to take on predicted loads.
The feature will be available initially in the US East (N. Virginia), US East (Ohio), US West (Oregon), Europe (Ireland), and Asia Pacific (Singapore) Regions.
He said, as well as improving users’ experiences, it should help cut users costs by avoiding over-provisioning. There will be no additional charges for the feature.