Amazon brings predictive scaling to EC2 instances

The update to Auto Scaling should help customers avoid over provisioning and reduce EC2 costs.
Written by Stephanie Condon, Senior Writer on

Amazon Web Services this week is improving its Auto Scaling tool with machine learning, giving it predictive capabilities. The new predictive scaling feature predicts a customer's expected traffic and EC2 usage, including daily and weekly patterns. It can help a customer create a scaling plan in anticipation of daily and weekly peaks.

The machine learning model behind predictive scaling is informed by data collected from a customer's own EC2 usage, as well as billions of other data points Amazon observes. The model needs historical data from at least one day to start making predictions. Every 24 hours, the model is re-evaluated to create a forecast for the next 48 hours.

Amazon EC2 was launched in 2006, while Auto Scaling was launched in 2009, along with CloudWatch Metrics and Elastic Load Balancing. These two launches "truly signify the fundamentally dynamic, on-demand nature of the cloud," Jeff Barr, chief evangelist for AWS, wrote in a blog post.

The update to Auto Scaling comes a week ahead of the annual AWS re:Invent conference. Last year's re:Invent conference was used to roll out a slew of new services, many powered by machine learning. Earlier this week, AWS announced updates to Amazon Polly, Transcribe and Translate, tools that use machine learning to improve customers' own products.

Editorial standards