Questionnaire for KubeRay benchmark
Hi all, the KubeRay team is planning scalability and memory usage benchmarks. Our goal is to ensure that KubeRay is stable enough to support more than 95% of use cases. However, as we currently do not have enough information on how users utilize KubeRay, we've decided to create a questionnaire to gather more data. Refer to kuberay#1208 for more details about benchmarking.

All data collected by this form will not be shared publicly. It will only be used by the Anyscale KubeRay team to determine the goals of the KubeRay benchmark.

If you have any questions, you can send an email to Kai-Hsun (kaihsun@anyscale.com) and cc Praveen (praveeng@anyscale.com) and Archit (archit@anyscale.com).
Sign in to Google to save your progress. Learn more
Name *
Email *
Company *
GitHub

How many KubeRay operators are there in your Kubernetes cluster? Is there one operator for the entire cluster? One operator per namespace? Or is there one operator for multiple namespaces?

Which CRDs do you use: RayCluster, RayJob, or RayService?

What’s the maximum number of custom resources (RayCluster, RayJob, or RayService) that an operator will handle at the same time?

What is the number of worker Pods in a single custom resource? It would be helpful if you could provide the 50th percentile (P50), 95th percentile (P95), and maximum (P99) values.

Do you use Ray Autoscaler? (https://github.com/ray-project/kuberay/blob/master/docs/guidance/autoscaler.md)
Do you use GCS fault tolerance? (https://github.com/ray-project/kuberay/blob/master/docs/guidance/gcs-ft.md)

How long does a RayJob typically take? It would be helpful if you could provide the 50th percentile (P50), 95th percentile (P95), and maximum (P99) values.

Submit
Clear form
Never submit passwords through Google Forms.
This form was created inside of Anyscale, Inc.. Report Abuse
Google Forms