Very Computer

Hello,

On a SunFire880 with Solaris 8 we are using Samba (2.2.8.1), in a
Prepress-department of a printing office, primarily to give our Windows 2000
RIPs access to the Suns raidvolumes.

For those who think: What's a RIP?
A RIP creates bitmap-files for a CTP imagesetter from Postscript- or
PDF-files. It creates lots of files, in several directories. These files are
mostly log-files, lowresolution- and highresolution-bitpmap files.

PROBLEM
When the RIPs process the Postscript/PDF-files, numerous files are created
in different directories on the Raidvolumes. All this happens mostly with
acceptable speed.
But when we try to delete 1 ripped job it takes the RIP about 180 seconds(!)
to delete it and continue the running tasks at normal speed. (For this 1
ripped job, actually about 400 files have to be deleted from the system.)
Considering that several jobs are deleted at the same time you can imagine
the long waits it creates.

BUT
When we try the same on a
- Windows 2000 shared directory or a
- Mac OS X.2 (with samba 2.2.3.b) shared directory
things work a lot faster.
The actual processing time can be up to twice as fast and deleting the job
takes about 3 to 5 seconds.
The performance of the network and Samba is okay, about 80-86% of FTP-speed.
On all systems.
All the configurations are reasonably standard and not tuned in special
ways.

QUESTION
How can it be that the combination Sun Solaris 8 and Samba 2.2.7 or 2.2.81
is so much slower than the other two configurations? Maybe someone knows
the differences between the mentioned configurations.
If there are settings that can speed things up I'd like to hear them!

smb.conf
==================================================
# Samba config file created using SWAT
# from localhost (127.0.0.1)
# Date: 2003/03/18 17:05:05
# Global parameters

[global]
workgroup = PREPRESS
netbios name = SUNFIRE
server string = Sun-Samba %v
interfaces = 192.168.10.10
security = SHARE
encrypt passwords = Yes
hosts equiv = /etc/hosts.equiv
log file = /usr/local/samba/var/log.%m
socket options = IPTOS_LOWDELAY TCP_NODELAY
dns proxy = No
veto files = /.*/Network Trash Folder/TheVolumeSettingsFolder/Temporary
Files/TheFindByContentFolder/FullPress:AA FPO/

[printers]
comment = All Printers
path = /usr/spool/samba
printable = Yes
browseable = No
available = No

[print]
path = /t3_1/orders/_Printmap
read only = No
create mask = 0777
directory mask = 0777
guest ok = Yes

[d4_store]
comment = Genstore Area op D4 voor Celebra RIP
path = /d4/d4_ripstore
read only = No
guest ok = Yes

[t3_store]
comment=Genstore Area op T3 voor Celebra RIP
path=/t3_1/t3_ripstore
read only = No
guest ok = Yes
==========================

Regards,

Henk Huitsing

I suggest that you enable logging on your ufs filesystem.  This will
_greatly_ speed up deletes, etc.  If you can, upgrade to Solaris 9,
update 2, where a lot more performance improvements have been incorporated.

You can generally do this on the fly with the remount option:

# mount -o remount,logging /export/t3_1   #... or whatever...

- Bart

        That is so long that it makes me think that the system is waiting
for something to time out.  (DNS?)  I assume that file deletions done
locally to the Solaris system don't take so long?  I would do a 'snoop' on
the network communications between one RIP and the Solaris host when you
try and delete a job; then you can see where the delay is.

--
If mail to me bounces, try removing the "+STRING" part of the address.