Re: more tcpdumpinfo for nfs3 problem: aix-server

Re: more tcpdumpinfo for nfs3 problem: aix-server

[Birger Lammering]

Hi,

still the same old amd/nfs3/aix-server - linux > 2.4.9 problem:
http://www.uwsg.indiana.edu/hypermail/linux/kernel/0111.2/0562.html

Now we've tried 2.4.18pre1 and got the same results as before. But
with Trond's patches things improved a bit:

patches applied:
pathconf, rpc_bkl, seekdir and ping (for 2.4.17)

results:
- no error messages while copying the file for the second time :-)
- the file was actually copied completely :-)
- it took veeery long :-(

qx09820@capc25 /home/qx09820 > time cp linux-2.4.9.tar.gz /scr/caes04/gaga3
0.020u 0.280s 0:07.61 3.9%      0+0k 0+0io 134pf+0w
qx09820@capc25 /home/qx09820 > time cp linux-2.4.9.tar.gz /scr/caes04/gaga4
0.010u 0.310s 6:17.28 0.0%      0+0k 0+0io 134pf+0w

This is still far from perfect, but something seems to happen...

Any ideas? More details?

Cheers,
Birger


--
Dr. Birger Lammering                         
science+computing ag                        fon: 089 356386-15
Geschaeftsstelle Muenchen                   fax: 089 356386-37
Ingolstaedter Str. 22
D-80807 Muenchen         mailto:B.Lammering@science-computing.de

Re: more tcpdumpinfo for nfs3 problem: aix-server

[Trond Myklebust]

>>>>> " " == Birger Lammering <b.lammering@science-computing.de> writes:

     > Hi,

     > still the same old amd/nfs3/aix-server - linux > 2.4.9 problem:
     > http://www.uwsg.indiana.edu/hypermail/linux/kernel/0111.2/0562.html

     > Now we've tried 2.4.18pre1 and got the same results as
     > before. But with Trond's patches things improved a bit:

Is this problem still only occurring when using amd?

Cheers,
   Trond

Re: more tcpdumpinfo for nfs3 problem: aix-server

[Trond Myklebust]

     > Strange situation: both, you and the amd-guys, think the bug is
     > somewhere in the code of the other.

Although I won't exclude the fact that there may be a kernel bug, as
long as this condition is not reproducible with anything other than
amd, I'm going to need a lot of convincing.

     > And I have no idea, how to circle in the
     > bug.... (Kernel-debugging???)

That depends a lot on what it is amd actually does. Is it for instance
acting as a proxy between the client and the server, or in any other
way touching those packets?
I've never used amd, so I frankly have no idea what sort of special
thing it does on top of the standard NFS client. That's why I'd prefer
that Ion look into it.

If it turns out that a proper bugreport can be compiled that points to
a kernel NFS client problem, I'll be happy to help fix it. As things
stand, however, I am unable to reproduce any such problems on my own
setups and I have no spare time at the moment to start messing with
amd myself.

Cheers,
  Trond

Re: more tcpdumpinfo for nfs3 problem: aix-server

[Birger Lammering]

Hi Trond,

Trond Myklebust writes:
 >      > Now we've tried 2.4.18pre1 and got the same results as
 >      > before. But with Trond's patches things improved a bit:
 > 
 > Is this problem still only occurring when using amd?


still the same: We couldn't reproduce the problem by mounting
manually. :-(

Strange situation: both, you and the amd-guys, think the bug is
somewhere in the code of the other. And I have no idea, how to circle
in the bug.... (Kernel-debugging???)

cheers,
Birger


 > 
 > Cheers,
 >    Trond
 >