gpu-manager icon indicating copy to clipboard operation
gpu-manager copied to clipboard

gaiaGPU对k8s 1.21版本兼容问题

Open ranxuxin001 opened this issue 1 year ago • 3 comments

ranxuxin001 avatar Jun 03 '24 02:06 ranxuxin001

大家好,我看到github上gaiaGPU的代码是2年前更新的。请问是否能兼容目前的k8s,或者有其他人基于目前k8s 1.21 版本能否成功部署gaiaGPU,实现GPU虚拟化和资源隔离。期待回复。谢谢。

ranxuxin001 avatar Jun 03 '24 02:06 ranxuxin001

@ranxuxin001 我们在1.28版本的k8s上成功调度起来了官方给的测试镜像,不过需要魔改代码,我们遇到的问题主要集中在对cgroup的魔改,当前的gpu-manager和vcuda-controller都是基于cgroup v1获取podUid,containerId和宿主机PID的,需要适配到cgroup v2上 image image

xxsoul avatar Jul 24 '24 02:07 xxsoul

We recently open-sourced another GPU Virtualization project, which not only supports GPU Virtualization, but also supports heterogeneous GPU management and flexible schedule.

HAMi -> https://github.com/Project-HAMi/HAMi, It is now a CNCF landscape project.

wawa0210 avatar Dec 05 '24 10:12 wawa0210