From: didier <dga...@ma...> - 2008-07-23 03:47:10
|
Hi, Le mardi 22 juillet 2008 à 19:24 +0200, Rolf Randen a écrit : > Hello again, > > now I got a capture of the described phenomenon, opened in wireshark - > but besides tons of checksum errors (yes, caused by "TCP checksum > offload") I don't understand too much. > > I tried to compare with a capture of a successful (file opened writable) > conversation: The only difference I could find at first view is a lot of > packets (server -> client) marked as "reply without query" in the > capture of the failed (readonly) conversation. And not a single one in > the "good" capture. Most of the time they are spurious errors because either: - You don't unset 'Validate TCP checksum ...' in TCP prefs and wireshark doesn't reassemble packets. - Tcpdump capture with truncated packets. - Wireshark/tcpdump didn't capture all packets but then you should get wireshark errors about missing packets. On the other hand you should get this error with the good capture too. > > Interesting for me is, that after I had changed the dbpath of the > volumes to another physical drive yesterday (hoping to improve the > perfomance), today two users without too much work have been enough to I'd expect it to improve speed too, but if it's a race faster means more problems not fewer. > cause that problem, while usually it happened only during the last > stressy hours of production (with about 50 users on the network, > including 10 users in the InDesign/InCopy workflow). > The cnid-dbs _now_ are in the root-fs (md_RAID-1/sata) while the > exported filesystems reside on a RAID-5 SCSI-subsystem (Ultra-320). > > Another new detail: The InDesign-files don't need to have objects > exported to the InCopy-workflow to be part of the problem. > > Any hints or ideas out there? Only the capture can help, if you can share the good and the bad one. Stuff to test: - If all users have the same local ID on their computer it could be a Word like bug (using the local user ID as unique key for temporary files in a share volume is not a good idea). - CS2 is rather old isn't it? IIRC first Apple OSX AFP servers didn't have a fully working files locking... It's possible to recompile netatalk without file locking but looking at the capture first would be better. Didier |