Why DC/OS, why mesos?
Abstraction of machines , not resources
Task scheduling and health
Let me introduce Apache Mesos and the concept of a datacenter kernel
Spring 2009: 262B project between BenH, Andy and Matei2009: Nexus project started at UC Berkeley’s RAD Lab (later renamed to Mesos)
March 2010: Grad students give talk at Twitter, Twitter starts exploring deployment
September 2010: Mesos technical report published (http://mesos.berkeley.edu/mesos_tech_report.pdf)
December 2010: Mesos enters Apache Incubator (http://incubator.apache.org/projects/mesos.html)
What are the main concepts important to understand in Mesos.
Mesos provides a level of indirection by being responsible of the allocation and reallocation of resources.
Weighted DRF (dominant resource fairness) and allocator module.
All of this messaging can seem a little over the top but let’s consider what happens if we have many more schedulers that want to do work and many agents to do that work. Now the master is essentially acting as a messaging hub - except that it has control over which schedulers get offered which resources and tries to allocate them fairly.
Marathon acts as the cluster “init” system
All additional services we launch via Marathon
Clustering your hardware this way allows you to share idle resources amongst applications and, more often than not, leads to some amount of cost saving through hardware reduction.
Mesos by itself is an abstraction of resources, by adding other tools on it we could use it as a paas
We started with mesos + marathon, very quickly more components started getting built such as logging,metrics, UI,universe, security for this to be enterprise production ready. We packed all of these components and built dc/os
Config parameters for cluster setup, networking options,