Re:Putting GPUs to work with Kubernetes
by eicnix on Wednesday April 19, 2017 @06:08AM
(#14146663)
You can run distributed stateful workloads on kubernetes with local storage and disable rescheduling when the node goes down. This means that you need to manually migrate/restart workloads if a node goes down. There is work being done on improving handling local node storage https://github.com/kubernetes/community/pull/306 but it doesn't scale as well as network storage.
If you have an existing SAN solution you can connect to it via fiber channel over iSCSI.
If you have an existing SAN solution you can connect to it via fiber channel over iSCSI.