> On 25 Apr 2021, at 15:06, Rick Macklem <rmacklem_at_uoguelph.ca> wrote: > > 2 - TSO does work correctly, but results in different timing of the TCP > segments transmitted for the segment compared with non-TSO. > > I believe that, for otis_at_, disabling TSO reduced the frequency of Linux > client hangs, but did not stop them. > --> reverting the patch in r367492 (this patch is not in FreeBSD12) has > fixed the problem for him. > Correct. Reverting the patch in r367492 has made the system stable and usable (thanks, Rick!). We also have disabled LRO and TSO on the interfaces serving the NFS traffic, it also might have added up to the stability. There is some more work going on in Phabricator (D29690) that we also want to test. otis — Juraj Lutter otis_at_FreeBSD.orgReceived on Sun Apr 25 2021 - 11:23:46 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:41:28 UTC