EECS Department Colloquium Series
Multi-tenancy and datacenters: to each according to his... 
 |
Wednesday, December 10, 2014 Ali Ghodsi
|
ABSTRACT:
More and more organizations are moving their applications to datacenters, where workloads are consolidated. These applications often have different requirements and constraints on the resources they need to use. A central conceptual question is: how much resources should be allocated to the different applications? In answering this, we consider the following three questions. First, can we guarantee that every application gets as much resources as if it ran in its own cluster? Second, can we ensure that applications cannot gain advantage over other applications by inflating their demands. Finally, can we ensure that organizational policies about resource allocation are satisfied when consolidating workloads. This talk outlines a framework for reasoning about these issues and covers one such policy, Dominant Resource Fairness (DRF), which now is the standard policy in two of the leading big data resource managers, Hadoop YARN and Apache Mesos.
BIOGRAPHY:
Ali Ghodsi has been a researcher at UC Berkeley, prior to which he was Assistant Professor at KTH/Royal Institute of Technology in Sweden. Ali is currently the Head of Engineering and PM at Databricks. His general interests are in distributed systems and networking.
| Return to EECS Joint Colloquium |