Skip to contentSkip to conversation

NFS dropouts with <=2.4.15pre1 + ReiserFS

DISCUSSION Jens Benecke 2 participants 3 messages Nov 11, 2001

lore
Share & tools
lore
#0 lore raw

Hi,

both to linux-kernel and reiserfs-list because currently I don't know who to blame. It should be related to ReiserFS + NFS however, because I can FTP between the systems and get 8-9MByte/sec.

I have two machines, ds9 (workstation) and server (well..), running 2.4.15pre1. The server was patched with grsecurity (www.grsecurity.net), the workstation was patched with Win4lin (commercial, www.netraverse.com).

As I said, FTP works fine. But when I copy something via NFS, performance is about a third to a fifth of the FTP performance - AND, from time to time (I'd say every 200-300MByte) NFS throughput drops out completely.

I turned on NFSd debugging and got lots and lots of "fh_verify", "fh_compose" etc messages in the kernel and found something very curious:

Whenever my iptables packet filter would log denied packets, NFS would drop out for a couple seconds to a couple minutes - then resume operations. My NFS client told me "nfs server not responding, NFS server OK" a couple times.

This seems reproducable with a little effort - so I'd be happy to help debugging this (and the bad performance ;) if anybody tells me where to start. As I said, with 'echo 9 > nfsd_debug' I got megabytes of logs filled with nfsd_dispatch, fh_verify, fh_compose, last message repeated 295 times, etc. but nothing that (for me) pointed to an error situation.

The same happened with all 2.4 kernels I tried so far, as far as I can remember.

Thanks in advance!

Signature · 1 line
Jens Benecke ········ http://www.hitchhikers.de/ - Europas Mitfahrzentrale
#1 lore raw
Quoted 3 lines · Jens Benecke

I have two machines, ds9 (workstation) and server (well..), running 2.4.15pre1. The server was patched with grsecurity (www.grsecurity.net), the workstation was patched with Win4lin (commercial, www.netraverse.com).

(...)

Quoted 5 lines

This seems reproducable with a little effort - so I'd be happy to help debugging this (and the bad performance ;) if anybody tells me where to start. As I said, with 'echo 9 > nfsd_debug' I got megabytes of logs filled with nfsd_dispatch, fh_verify, fh_compose, last message repeated 295 times, etc. but nothing that (for me) pointed to an error situation.

First advice is to reproduce this without external patches.

Regards,

bert

Signature · 4 lines
http://www.PowerDNS.com          Versatile DNS Software & Services
Trilab                                 The Technology People
Netherlabs BV / Rent-a-Nerd.nl           - Nerd Available -
'SYN! .. SYN|ACK! .. ACK!' - the mating call of the internet
#2 lore raw
In reply to Jens Benecke
Quoted 9 lines · bert hubert
Quoted 6 lines · Jens Benecke

This seems reproducable with a little effort - so I'd be happy to help debugging this (and the bad performance ;) if anybody tells me where to start. As I said, with 'echo 9 > nfsd_debug' I got megabytes of logs filled with nfsd_dispatch, fh_verify, fh_compose, last message repeated 295 times, etc. but nothing that (for me) pointed to an error situation.

First advice is to reproduce this without external patches.

Will do. (as soon as I get a chance)

Signature · 5 lines
Jens Benecke ········ http://www.hitchhikers.de/ - Europas Mitfahrzentrale
   · . ·
· · . · . ·  <-------- verdächtiges weisses Puder
 · . · . .
    · ·