AWS Auto Scaling by Appliscale

Some details

One of the most meaningful players in the US advertising market runs an AdTech platform that supports billions of request per second.
The whole infrastructure is serverless, hosted in the cloud environment. Such high traffic requires a tremendous amount of computing resources, which generates million-dollar bills.
To cut monthly invoice costs, Appliscale utilized AWS Auto Scaling feature. Analyzing the most recent

… traffic statistics and long term weekly observations, our engineers were able to extract traffic patterns and create a dynamic scaling policy.
 

EC2 machines are scaled out according to the most recent average requests volume and the time of the day. A number of nodes is always balanced so that incoming traffic is handled by a minimal reliable hosts amount.

Another consequence is auto-healing infrastructure - healthy hosts replace unhealthy ones without any manual intervention. Subsequently, the infrastructure becomes immutable - the easiest way to deal with the misbehaving host is to tear it down - and automation takes care of anything besides.

Thanks to ongoing metrics analysis using DataDog Appliscale engineers implemented tailored scaling policy.

Using automation tools such as CloudFormation, Ansible, and Jenkins, to name a few, we designed and developed a simplified and more efficient release process.

Creating a dynamic scaling policy and fully immutable infrastructure have reduced costs of EC2 infrastructure maintenance by roughly 40% monthly, which resulted in significant savings for the client.