MesosCon NA 2016

The Oracle Data Cloud is processing billions of events and multiple terabytes of data per day in real-time. In this presentation Mike Sells will discuss the motivation, design, and implementation of an elastic Mesos platform to handle the needs of their real-time systems. They will discuss how to think about the elasticity of applications inside of Mesos as well as Mesos itself with details on how to implement both. Why scaling down, not up, is the biggest challenge. They will explain how they built a high volume low latency web tier and a Spark streaming data pipeline on top of the same self-healing platform. How building this multi-tenant system on top of Mesos can maximize resource utilization as well as productivity while automatically adjusting to the demands of traffic and self healing on failures. This includes how they test and validate that elasticity does not impact data integrity. And will conclude by demonstrating the real-time production system, creating live failures, and showing its ability to self heal and auto correct.