Please log in to watch this conference skillscast.
Firmament models the scheduling problem as an optimisation over a flow network. This approach considers all possible assignments concurrently and makes placement decisions that are optimal for the given scheduling policy. Ionel illustrates how this enables practical, useful scheduling policies to be expressed concisely. With the aid of two case studies, you will discover that Firmament makes high-quality decisions, reducing the per-task runtime by 2-4x over the widely-deployed Mesos cluster manager’s default decisions. In order to scale to large clusters, Firmament solves the optimisation problem incrementally. On a Google cluster of 12,500 machines, it makes decisions in 200ms on average, sufficiently fast for interactive applications.
Firmament is an Apache-licensed open-source project and includes both a cluster manager and several schedulers, and we are working on adding support for using Firmament atop Kubernetes and Mesos.
Find out more about Firmament
YOU MAY ALSO LIKE:
Accurate, flexible and scalable scheduling with Firmament
Ionel Gog is a PhD student at Cambridge and works in distributed systems and algorithms. In the past, he has interned in the Borg team and YouTube mobile teams at Google and in the Data Infrastructure team at Facebook.