You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While running some high-performance benchmark of a multithreaded application on a 100GE client,
we noticed, that although our storage endpoint can deliver 20 GiB/s, a single application was limited to ~2.5 GiB/s data input (single multithreaded application running 85% idle) (streaming many files in parallel)
After some investigation, it turned out, that setting:
env XRD_PARALLELEVTLOOP=16
allowed to run this application at 5.8 GiB/s
This ticket is about the possibility of some automatic configuration of the eventloop size in XrdCl to avoid tweaking the settings for each environment.
The text was updated successfully, but these errors were encountered:
We've just increased the default value of XRD_PARALLELEVTLOOP to 10, which should be enough in most situations. Should we proceed to close this ticket, or is there still the need to have a better approach which dynamically sets the number using some heuristics?
While running some high-performance benchmark of a multithreaded application on a 100GE client,
we noticed, that although our storage endpoint can deliver 20 GiB/s, a single application was limited to ~2.5 GiB/s data input (single multithreaded application running 85% idle) (streaming many files in parallel)
After some investigation, it turned out, that setting:
env XRD_PARALLELEVTLOOP=16
allowed to run this application at 5.8 GiB/s
This ticket is about the possibility of some automatic configuration of the eventloop size in XrdCl to avoid tweaking the settings for each environment.
The text was updated successfully, but these errors were encountered: