On 2013-01-25, Grant Edwards <grant.b.edwa...@gmail.com> wrote: > I've been examining Ethernet traces from a customer site where there > appear to be problems associated with the eCos ARP table entry > timeouts.
It seems this is a known problem with a known fix, so I thought I would follow-up just for the record... Bernd Edlinger has kindly pointed me to a patch he submitted that among other things fixes the periodic packet loss due to ARP entry timeouts: http://bugs.ecos.sourceware.org/show_bug.cgi?id=1001656 It would be interesting to have a chat with whoever decided that letting the entry expire every 20 minutes [and thus potentially discarding tx packets] was the right thing to do. While it's obvious that discarding eCos's tx packets is part of the design of the BSD ARP subsystem, what I haven't figured out is why the TCP connection doesn't recover. The smart money is betting on Microsft as the culprit: one end of the TCP/IP connection is Windows 7, and the customer claims they never had the problem when using XP. How long as TCP/IP been around? It seems nothing is so old and well understood that it can't be screwed up by Microsoft... -- Grant -- Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
