clipper
clipper copied to clipboard
[Performance] Performance Degradation on Kubernetes
User is reporting:
I set up clipper on a Kubernetes cluster and tried to see how models perform. I found that the response time of models (even the helloworld example on clipper homepage) kept increasing as I submitted more queries.
I will try to re-produce this.
I will try to re-produce this.
Thanks. I tried on both minikube and a dedicated Kubernetes cluster, and the issue happened on both case.
I used the tutorial model (sum-list), With the following data input in every query:
data = [random.uniform(0, 500) for _ in range(2)]
Hi, @simon-mo How's the status for this?