Abstract: |
|
This session is covering OpenStack and Ironic (bare metal as a service) for creating HPC clusters, on demand, either virtual, or physical, or even a mixture of both — in mere minutes. At Bright Computing we operate our own OpenStack private cloud, and Ceph cluster, to power it. We make heavy use of OpenStack Heat and Ceph's copy on write (from glance to cinder) to enable the end users of our OpenStack cloud to easily (and quickly!) create entire complete clusters from scratch, within, literally, minutes. This is a solution we came up with and called it "Cluster on Demand" (COD). COD clusters can be easily used for anything: HPC, Hadoop, Spark, Deep Learning, etc. They are already preprovisioned and preconfigured with all the necessary packages, libraries, compilers, workload managers, as well as customized out of the box to users requirements. Moreover, COD clusters are not restricted to VMs. They can also run on physical clusters by leveraging OpenStack’s “Bare metal as a service” component called Ironic. And the best part about it: the end user who is creating the clusters enjoys the same user experience regardless of whether it's a virtual cluster (e.g. experimenting, trying out different configurations), or physical cluster (hardcore, zero-overhead, HPC). Join Bright Computing CTO Martijn de Vries for a lively tutorial on how bare metal (or virtual) HPC Clusters can run on Demand driving higher efficiency across your clustered IT infrastructure while giving a sysadmin an opportunity to enjoy managing it. |
|