The main idea behind the improvement is to add NUMA node id of GPU card to *pluginapi.Device. This information can be used by TopologyManager of kubelet to allocate GPU devices along with other devices on the same NUMA node, such as a another GPU card or set or CPU.
t33m
commented
9 months ago
The PR adds tests to previously created by @silenceper PR https://github.com/NVIDIA/kubevirt-gpu-device-plugin/pull/76.
The main idea behind the improvement is to add NUMA node id of GPU card to
*pluginapi.Device. This information can be used by TopologyManager of kubelet to allocate GPU devices along with other devices on the same NUMA node, such as a another GPU card or set or CPU.More info about