Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Performance tests on CESNET infra #16

Open
guillaumeeb opened this issue Sep 15, 2022 · 6 comments
Open

Performance tests on CESNET infra #16

guillaumeeb opened this issue Sep 15, 2022 · 6 comments

Comments

@guillaumeeb
Copy link
Member

In #15, some first tests on using Xarray on CESNET with Swift are proposed. This shows that (thanks to @sebastian-luna-valero), it's quite easy (even with some manual steps) to write on the Swift object storage using Zarr and Dask.

One thing we might want to do know would be to perform some light but comprehensive benchmarks to identify what performances we could get on this Infrastructure.

A classical benchmark could be:

  • Define some example datasets: small, medium, large (up to some TiB?).
    • 10GiB
    • 100GiB
    • 1TiB
  • Write these datasets on Dask clusters varying in size:
    • 5 workers
    • 10 workers
    • 20 workers
  • Read back the datasets on varying Dask clusters
  • Compute things like troughput or other stats (maybe by analyzing the Dask task report)
  • Analyze the results.

We might want to start from something like https://github.com/pangeo-data/benchmarking on wihch @tinaok contributed.

We need also to ask CESNET for potential limits or constraints they'd have @sebastian-luna-valero.

@sebastian-luna-valero
Copy link
Collaborator

Sounds like a good plan to me, thanks!

I am asking CESNET about the potential limits and constraints are will report back.

@sebastian-luna-valero
Copy link
Collaborator

According to CESNET benchmark tests can start right away. We have 10TB quota for object storage.

They shared this list of flavors: https://docs.cloud.muni.cz/cloud/flavors/

We are currently using the hpc.16core-64ram-ssd-ephem flavor with an average network throughput of 1250.0 Mbps. If that's not enough, we should switch flavors to one without limits.

@guillaumeeb
Copy link
Member Author

If there is no limit, except the volume, let's try to scale :). We'll see later if we need bigger throughput.

@guillaumeeb
Copy link
Member Author

So I've opened pangeo-data/benchmarking#45 if you want to look at it.

@sebastian-luna-valero
Copy link
Collaborator

Should we close this one, or redo it with MinIO?

@guillaumeeb
Copy link
Member Author

I think we can leave it open, although I'm not sure I'll work on it soon... But this would give us some figures of the performance we can reach with our infrastructure!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants