Practical Container Scheduling: Juggling Optimizations, Guarantees, and Trade-Offs at Netflix -

Practical Container Scheduling: Juggling Optimizations, Guarantees, and Trade-Offs at Netflix - Sharma Podila, Netflix

Assigning resources from distributed clusters to containers from disparate use cases can be tricky. It becomes even more so in practice when combining multiple scheduling objectives and constraints such as bin packing, task locality, and capacity guarantees, among others.

Containers are increasingly used to run microservices, batch, and stream processing applications in large scale shared Mesos clusters at Netflix. This talk dives deep into the challenges, design, and trade-offs achieved using open source scheduling library, Fenzo, that takes a holistic approach, along with plugins based extensibility, to provide a nimble a scheduling core for various independently evolving clusters. Using results, we highlight aspects of capacity guarantees, task placement, elasticity, and operational insights to tackle large scale operations.


Sharma Podila
Netflix
Senior Software Engineer
Los Gatos
Websitenetflix.com/
Sharma Podila is a Senior Software Enginner in the Edge Engineering team at Netflix, Inc. His current work includes Fenzo, an open source generic scheduler with plugin based optimizations, developing cloud native Mesos frameworks, and evolving the microservices platform at Netflix.