What metrics are used by load based auto scaling?

Load based auto scaling currently allows you to define thresholds for the following metrics. A breach of any metric will result in scaling up for upscaling thresholds and scaling down for downscaling thresholds.

• CPU: The average CPU consumption over all instances in the role, e.g. 80.5 %
• Memory: The average memory consumption over all instances in the role, e.g. 60 %
• Load: The average 1-min load over all instances in the role, e.g. 3.5

You can define over how many measure periods the averages should be computed. Instances with not enough metrics will be ignored. Freshly booted instances will be ignore for a couple of minutes.