2016 OpenStack Barcelona – Gnanavelkandan Kathirvel, Chip Kelley – Scaling OpenStack to large scale
OpenStack has known scalability limitations and scaling beyond 100-200 compute nodes per single OpenStack instance (Region) is a challenge. Scaling out to support 500 + nodes in a single datacenter is possible by deploying multiple OpenStack instances in that location but that adds extra overhead, cost and management complexity for tenants.
In this case study, we will share our performance and scale test results, fine tuning that ‘worked’ versus ‘did not work’ to support the scale and the architecture changes made to OpenStack deployment to support 500+ compute nodes per OpenStack region.
We have addressed bottlenecks in the messaging bus, Nova scheduler, and other key components, with a combination of:
● Control plane flexibility (moving services to different hosts)
● Configuration changes particularly with RabbitMQ and Nova
● Tuning the use of Ceilometer
by vBrownBag
openstack
