We slashed our Amazon DynamoDB costs by over 75% using Kinesis, DynamoDB streams, and Erlang/OTP to implement a global cache warming system for advertising profile data. Using Elixir and the Flow framework, we subsequently doubled the performance of that system with only minor modifications. This talk will describe our motivation, some pitfalls we encountered, and how use of Erlang and now Elixir ultimately led to the success of this project. We will also present some examples for doing this kind of scalable high-volume processing (up to 2.5 million updates/sec globally in our case) using two new open-source libraries from AdRoll.
Mike Watters, is a staff engineer at AdRoll where he has been for 3 and half years working on improving the real time bidding infrastructure.
Read more at code sync web page