how to test the runtime isolation of computing resource and memory？

NTHU-LSALAB / KubeShare

Share GPU between Pods in Kubernetes

Apache License 2.0

193 stars 42 forks source link

how to test the runtime isolation of computing resource and memory？ #5

Closed koalaxiu7 closed 3 years ago

koalaxiu7 commented 4 years ago

hi, I want to test the runtime isolation of computing resource and memory, how to test the isolation and where are testcases for it? Thanks in advance!

jchou-git commented 3 years ago

We used detectron2 inference benchmark. To control the workload, we adjust the inter-arrival time of inference requests. For instance, you can increase the workload by reducing the inter-arrival time to force higher GPU utilization. By with our GPU isolation mechanism(Gemini), the GPU utilization should be bounded under the limit specified in YAML.