sharedgpu/group_threshold (optional): the minimum proportion of pods to be scheduled together in a pod group.sharedgpu/group_headcount (optional): the total number of pods in same group.sharedgpu/pod_group_name (optional): the name of pod group for a coscheduling.priority is greater than 0 represented as Guarantee Pod, which optimizes performance considering locality.priority is equal to 0 represented as an Opportunistic Pod used to defragmentation.sharedgpu/priority(optional): pod priority 0~100. sharedgpu/gpu_mem (optional): maximum GPU memory usage of Pod, in bytes.sharedgpu/gpu_limit (required if allocating GPU): maximum extra usage if GPU still has free resources, gpu_request sharedgpu/gpu_request (required if allocating GPU): guaranteed GPU usage of Pod, gpu_request It can't compatible with other scheduler to manage gpu resourceīecause floating point custom device requests is forbidden by K8s, we move GPU resource usage definitions to Labels.You also ensures that the prometheus is installed, because we will pull the data from it.Only support a kubernetes cluster that uses the environment variable NVIDIA_VISIBLE_DEVICES to control which GPUs will be made accessible inside the container. SharePod 4.1.0 registration code is a very easy application no need for installation simply download open and enjoy the features of sharepod.A Kubernetes cluster with garbage collection, DNS enabled nvidia-continaer-runtimeinstalled.Support GPU Heterogeneity & Topology awareness.Refer to the KubeShare 1.0 branch for the old version. KubeShare 2.0 is designed in the way of the scheduling framework. KubeShare □ □ Kubeshare 2.0 is now avaible, version 1.0 will be deprecatedĪ topology and heterogeneous resource aware scheduler for fractional GPU allocation in Kubernetes cluster
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |