On december last year, the last collisions were recorded on the 4 experiments to close a very successful LHC second running period (Run 2). At this very moment, the collaborations at CERN are working hard to upgrade the LHC for the next run that will start in 2021. During the next run, the computational needs required to process all the data produced in the LHC will increase dramatically.
In order to follow these upcoming needs, the CERN private cloud is always looking into ways of optimizing and make more resources available to our clients. We’ll review the tools that allows us automate the distribution of workloads, increase their efficiency and optimize the cloud environment. This tools are based in OpenStack services like Mistral and Watcher. Finally we’ll show our upcoming work in to push even further the limits of our service offering by looking into optimizing workloads in Kubernetes clusters and preemptible instances.
The cloud abstraction provides the illusion of infinite resources available for the end user to consume. Then, every cloud service needs to follow the utilisation trend and make resources available before the end user actually demands it. However, adding resources to the pool, sometimes it is not easy or it is tied to business constraints. This makes the equation even more complex.
By optimizing the resources, the cloud provider gets the most from the infrastructure it manages.This provides flexibility to the cloud provider to steer its infrastructure to meet user demand.
Attendees will learn techniques that we are currently using to make more resources available to our clients. Such techniques comprise consolidate or distribute workloads to maximize CPU/IO utilisation, clean up unused resources, and automatic optimization of workloads. These techniques are using existing OpenStack services like Mistral and Watcher.