Mesos master Mesos slave Mesos slave decides how to allocate resources service Y scheduler service Y task (Netty server) service Y executor Netty Server
Mesos master Mesos slave service X executor Mesos slave task launches, isolates, and monitors tasks and executors service Y scheduler service Y task (Netty server) service Y executor Netty Server request offer
managing productions servers easy jobs/servers are submitted to the framework via a configuration file provides mechanisms: » rolling restarts/updates » relaunching processes after failures (if requested) » and more!
actor style concurrency (github.com/libprocess) APIs in C++, Java, Python protobuf for data transport, data types zookeeper support for high-‐availability linux control groups support (LXC/cgroups)
new framework for job workflows, wants to use Spark and Hadoop too Built DPark (a Python clone of Spark), also running MPI Hadoop and Spark used by machine learning researchers