[OpenAFS] Re: client hangs with 1.2.13 redhat 9 servers

[Mike Fedyk]

Jeffrey Hutzelman wrote:

>
>
> On Friday, February 04, 2005 01:58:06 PM -0500 Joe Buehler 
> <jbuehler@spirentcom.com> wrote:
>
>> A further point of information.  Using tcpdump on the main
>> two fileservers I see a lot of udp checksum errors.
>> One of the machines has fragment reassembly timeouts, which
>> probably explains the AFS delays.
>>
>> I am a bit puzzled by tcpdump claiming errors in the UDP
>> checksum for *outgoing* packets on one of the machines,
>> however.  At what point does tcpdump hook into the
>> ethernet hardware?  Seems like UDP checksums in outgoing
>> packets should be impossible...
>
>
> I've noticed that too in some cases, and not just with AFS.
> It seems that in at least some situations, tcpdump does not get to see 
> the real UDP checksum that will appear on the wire for outgoing 
> traffic.  This is entirely dependent on how the operating system 
> implements its packet tracing facility.

Does your ethernet card have hardware checksumming?  Maybe tcpdump is 
getting the packets before the hardware has put in the checksums...