Fredrik Pettai <pettai%nordu.net@localhost> writes: > On Jan 29, 2010, at 6:08 PM, Andreas Gustafsson wrote: >> Matthias Scheler wrote: >>> Are you using hardware checksum offload on that "re0" interface? >>> If you do please try to disable it and reenable SACK. >> >> My hardware, network topology, and NetBSD version have all changed >> since 2008, and now when I try to reproduce the problem, I find that I >> am unable to do so. Apparently whatever was broken has been fixed >> either in NetBSD or by the hardware changes. I have now re-enabled >> SACK on all my NetBSD systems, and if the problem recurs, I will file >> a PR. > > Before I forget to ask, which NetBSD version are you using with a > functioning SACK? > We had problems with SACK on NetBSD 5.0_RC... something, > not sure that we ever came to test 5.0_RELEASE or later... SACK is hairy and I can certainly believe there are bugs. But I have looked at tcpdump traces (with tcpdump2xplot) and seen sack blocks and retransmits that look ok. I am still a bit suspicious of whether the choice of how much and what to retransmit under multiple losses is exactly correct. But I can't make an argument, even incoherent, that it's wrong.
Attachment:
pgp91pqXoA1CW.pgp
Description: PGP signature