A GARP or two later: Managing VMs in a busy OpenStack deployment

For optimal reading, please switch to desktop mode.

Baremetal compute is synonymous with high performance computing, and for good reason. Nobody wants their matrix multiplication running through a sluggish software abstraction. And for this reason, virtual machines are often overlooked for compute intensive tasks. With modern hardware offload and virtualisation capabilities, this no longer needs to be the case. A virtual machine can exist anywhere on the performance continuum between baremetal and full emulation.

With virtualisation comes great flexibility. High performance hardware can be split between users. In the world of AI and trillion transistor boxes, compute resource can trivially shift between training and inference. At the click of a button, workloads can be migrated to anywhere in the world.

So how does it work in practice? Learn more about our experiences, working together with G-Research, in a-state of the art OpenStack deployment.

StackHPC

Other articles

OpenInfra Live: Ironic in Production

Kubeflow on Baremetal OpenStack