English version|中文版
HAMi, formerly known as 'k8s-vGPU-scheduler', is a Heterogeneous device management middleware for Kubernetes. It can manage different types of heterogeneous devices(like GPU,NPU,etc...), share heterogeneous devices among pods, make better scheduling decision based on topology of devices and schedule policies.
It aims to remove the gap between different Heterogeneous devices, and provide a unified interface for user to manage with no change to your application. Until June 2024, HAMi has been widely used around the world at a variety of industries such as Internet/Cloud/Finance/ Manufacturing. More than 40 companies or institutions are not only end users but also active contributors.
HAMi is a sandbox and landscape project of
Cloud Native Computing Foundation(CNCF),
CNAI Landscape project.
HAMi provides device virtualization for several heterogeneous devices including GPU, by supporting device sharing and device resource isolation. For the list of devices supporting device virtualization, see supported devices
- Allows partial device allocation by specifying device memory.
- Imposes a hard limit on streaming multiprocessors.
- Permits partial device allocation by specifying device core usage.
- Requires zero changes to existing programs.
A simple demostration for device isolation: A task with the following resources.
resources:
limits:
nvidia.com/gpu: 1 # requesting 1 vGPU
nvidia.com/gpumem: 3000 # Each vGPU contains 3000m device memory
will see 3G device memory inside container
HAMi consists of several components, including a unified mutatingwebhook, a unified scheduler extender, different device-plugins and different in-container virtualization technics for each heterogeneous AI devices.
The list of prerequisites for running the NVIDIA device plugin is described below:
- NVIDIA drivers >= 440
- nvidia-docker version > 2.0
- config default runtime is nvidia for containerd/docker/cri-o container runtime.
- Kubernetes version >= 1.16
- glibc >= 2.17 & glibc < 2.3.0
- kernel version >= 3.10
- helm > 3.0
First, Label your GPU nodes for scheduling with HAMi by adding the label "gpu=on". Without this label, the nodes cannot be managed by our scheduler.
kubectl label nodes {nodeid} gpu=on
Add our repo in helm
helm repo add hami-charts https://project-hami.github.io/HAMi/
Use the following command for deployment:
helm install hami hami-charts/hami -n kube-system
Customize your installation by adjusting the configs.
Verify your installation using the following command:
kubectl get pods -n kube-system
If both vgpu-device-plugin
and vgpu-scheduler
pods are in the Running state, your installation is successful. You can try examples here
HAMi-WebUI is available after HAMi v2.4
For installation guide, click here
Monitoring is automatically enabled after installation. Obtain an overview of cluster information by visiting the following URL:
http://{scheduler ip}:{monitorPort}/metrics
The default monitorPort is 31993; other values can be set using --set devicePlugin.service.httpPort
during installation.
Grafana dashboard example
Note The status of a node won't be collected before you submit a task
- If you don't request vGPUs when using the device plugin with NVIDIA images all the GPUs on the machine may be exposed inside your container
- Currently, A100 MIG can be supported in only "none" and "mixed" modes.
- Tasks with the "nodeName" field cannot be scheduled at the moment; please use "nodeSelector" instead.
The project is governed by a group of Maintainers and Committers. How they are selected and govern is outlined in our Governance Document.
If you're interested in being a contributor and want to get involved in developing the HAMi code, please see CONTRIBUTING for details on submitting patches and the contribution workflow.
See RoadMap to see anything you interested.
The HAMi community is committed to fostering an open and welcoming environment, with several ways to engage with other users and developers.
If you have any questions, please feel free to reach out to us through the following channels:
- Regular Community Meeting: Friday at 16:00 UTC+8 (Chinese)(weekly). Convert to your timezone.
- Email: refer to the MAINTAINERS.md to find the email addresses of all maintainers. Feel free to contact them via email to report any issues or ask questions.
- mailing list
- slack | Join
Link | |
---|---|
CHINA CLOUD COMPUTING INFRASTRUCTURE DEVELOPER CONFERENCE (Beijing 2024) | Unlocking heterogeneous AI infrastructure on k8s clusters Starting from 03:06:15 |
KubeDay(Japan 2024) | Unlocking Heterogeneous AI Infrastructure K8s Cluster:Leveraging the Power of HAMi |
KubeCon & AI_dev Open Source GenAI & ML Summit(China 2024) | Is Your GPU Really Working Efficiently in the Data Center?N Ways to Improve GPU Usage |
KubeCon & AI_dev Open Source GenAI & ML Summit(China 2024) | Unlocking Heterogeneous AI Infrastructure K8s Cluster |
KubeCon(EU 2024) | Cloud Native Batch Computing with Volcano: Updates and Future |
HAMi is under the Apache 2.0 license. See the LICENSE file for details.