GPU Sharing with HAMi

Mon, 01 Jan 0001 00:00:00 +0000

HAMi (Heterogeneous AI Computing Virtualization Middleware) is a CNCF Sandbox project that enables fractional GPU sharing in Kubernetes. Instead of dedicating an entire GPU to a single workload, HAMi lets containers request specific amounts of GPU memory and compute cores.

This guide covers GPU sharing for containers in tenant Kubernetes clusters. For GPU passthrough to virtual machines on the management cluster, see GPU Passthrough.

How it works

HAMi sits between the Kubernetes scheduler and the NVIDIA GPU driver:

How to relocate etcd replicas in tenant clusters

Mon, 01 Jan 0001 00:00:00 +0000

Tenant Kubernetes clusters are using their own etcd clusters, not the one that is used by the management cluster. Such etcd clusters are deployed in tenants and are available to managed Kubernetes clusters deployed in the tenant and its sub-tenants.

Replicas of a tenant etcd cluster can be relocated between nodes for maintenance reasons. Currently, management operations for tenant etcd clusters are not automated, but such a task can be done manually.

Managed (Tenant) Kubernetes on Cozystack

GPU Sharing with HAMi

How it works

How to relocate etcd replicas in tenant clusters