Open ivukotic opened 3 weeks ago
Could you please elaborate on what you mean? Have you tried using the recorder plugin for the client? I am thinking about creating some way to easily plot the output coming from that plugin, but I'm not sure this is what you need. Cheers,
Hi, We're building an XRootD testing setup at Edinburgh and may be able to contribute to this, if there's interest?
We're primarily working on setting up a multinode test setup for functionality testing to support UK sites. Ideally we would like to provide a platform to reproduce any problems reported by a UK XRootD site.
(Using k8s as a deployment platform partly as an excuse to build some familiarity within our site for more advanced container orchestration strategies/management).
I've been eyeing up using this to build a reproducible CI giving us "average performance vs X" between builds, if there's interest in this please let us know.
I have never heard about the recorder thing. Can you please point me to some documentation?
What I would really want is that each time there is a new build, a new series of xcache performance tests is performed. These tests should vary the way data is read (xrdcp, 90%, 50%, 30%, 10%), block size, latency to origin, just to mention a few. Plots of these results should be provided on the xrootd page. Also evolution of these values over time.
This is an interesting idea to be discussed in the upcoming XRootD and FTS Workshop. I would be very happy to have such tests as well. There are some tests ongoing with EOS in the context of RNTuple as well, it's a collaboration between CERN IT and EP-SFT. Some of these tests will involve xcache.
As somebody who has to tune xcaches for optimal performance, I would very much appreciate if every xrootd release would come with a set of curves showing throughput rate vs block size for both xrdcp and "analysis like" accesses.