[OpenAFS] file corruption redux

Miles Davis miles@CS.Stanford.EDU
Wed, 31 May 2006 10:00:39 -0700


On Fri, May 26, 2006 at 03:25:59PM -0400, Derrick J Brashear wrote:
> On Fri, 26 May 2006, Miles Davis wrote:
> 
> >I think I can get it to happen pretty regularly on several clients. It 
> >seems to
> >be just this server, though...I'm still stumped. Another point of possible
> >interest -- I see no corruption if I go through an afs client on the server
> >itself.
> 
> So is it possible that hardware is in some (non-disk-controller) way 
> corrupting the transfers?

Sure, I suppose, but I can't think of what could do it -- cpu/cache? Magical 
corrupting ethernet interface or driver (intel e1000)?

> It might be worthwhile to tcpdump -s (larger than mtu) -w (some file) host 
> (a client you know will lose) and port 7000 and then given cmp -l output 
> of the good and corrupt file perhaps we can see if there's garbage on the 
> wire.

OK, here we go: cmp -l aspell-bg-0.50-9.i386.rpm /tmp/aspell-bg-0.50-9.i386.rpm
 1683429 377 177

(same from at least two clients)

tcpdump (4.4MB) file is at http://cs.stanford.edu/people/miles/tcpdump.out
Server is 171.64.64.67, client is 171.64.64.132.

-- 
// Miles Davis - miles@cs.stanford.edu - http://www.cs.stanford.edu/~miles
// Computer Science Department - Computer Facilities
// Stanford University