GNOME Bugzilla – Bug 761812
netclientclock: Only use observations with a RTT smaller than the median
Last modified: 2018-11-03 12:32:53 UTC
See commit message
Created attachment 320778 [details] [review] WIP: netclientclock: Only use observations with a RTT smaller than the median The idea is that observations with lower RTT are more accurate. Use a sliding window of 64 observations but only take those for the linear regression that have a RTT smaller than the median. This hopefully simplifies this code a lot while at the same time improving the clock accuracy. Previously we had complicated filtering logic that was based on a running average RTT, median of the last 9 observations and discontinuity checking for the applied observations. This should be unnecessary now. This needs careful testing under various scenarios.
It might also make sense to split the window in two or three parts, and take the below-median delays of those 2/3 bins then. That way we will always consider measurements over a longer time, instead of possibly just from the beginning or end of the window.
What's up with this?
Comment on attachment 320778 [details] [review] WIP: netclientclock: Only use observations with a RTT smaller than the median It was better in some cases, worse in others. Definitely needs more work and proper analysis.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/gstreamer/gstreamer/issues/155.