NVIDIA 的文章讨论了共享 GPU 平台中的租户架构,认为为每个团队单独建 Kubernetes 集群往往过于严格,提出通过隔离租户集群在共享资源下实现隔离与资源利用之间的平衡。
Running a dedicated Kubernetes cluster per team often results in more isolation than an organization requires. While one cluster can be successfully shared...