samba - Mar 2006 - print server performance issues

I'm migrating a 1.000 queues windows print server with AD very critical
enviroment to linux/cups/samba solution. Initially, everything working fine, but
when reached paroximately 400 queues created the server is getting degradated.
The browsing of printers shares is very low. Many times during the day smbd and
winbindd need to be restarted to stop the degradation. I didn't found any
bottleneck, memory, processor, disk i/o are fine. I have read a lot of tuning
docs but nothing solved my problem. Some body knows where can I find more
information? Or if there's any study case with 1000 queues in a enviroment
with
10.000 workstations and aproximately 20.000 users registered in MS-AD. Is samba
prepared to this kind of enviroment?

Thanks in advance,

Bruno Gomes Pessanha

Using:
Red Hat Enterprise Linux 4
2.6.9-34.ELsmp
samba-3.0.10-1.4E.6
cups-1.1.22-0.rc1.9.10

[root@tambaqui samba]# cat /etc/samba/smb.conf
# Global parameters
[global]
        workgroup = BC
        server string = TAMBAQUI Server
        security = domain
        password server = dc01, dc02, dc03, dc04, dc05
        log file = /var/log/samba/%m.log
        max log size = 50
        socket options = TCP_NODELAY SO_RCVBUF=8192 SO_SNDBUF=8192
        printcap name = /etc/printcap
        dns proxy = No
        wins server = 10.163.200.101, 10.163.200.102
        remote announce = dc01, dc02, dc03, dc04, dc05
        remote browse sync = dc01, dc02, dc03, dc04, dc05
        idmap uid = 10000-1000000
        idmap gid = 10000-1000000
        winbind separator = +
        winbind cache time = 3600
        cups options = raw
        load printers = yes
        show add printer wizard = no
        printer admin =
@BC+GG_BC_TESTE_PRINTER,BC+jgb9,BC+rsao,BC+cm9x,BC+ascm9x
        printcap cache time = 60
        client schannel = no
        lprm command = /usr/bin/cancel %p-%j
        lppause command = /usr/bin/lp -i %p-%j -H hold
        lpresume command = /usr/bin/lp -i %p-%j -H resume
        queuepause command = /usr/bin/disable %p
        queueresume command = /usr/bin/enable %p
        deadtime = 5
        write cache size = 262144


[printers]
   comment = All Printers
   path = /var/spool/samba
   browseable = no
   guest ok = yes
   writable = no
   printable = yes

[print$]
   path = /etc/samba/drivers
   public = yes
   writable = no
   browseable = yes
   write list = @BC+GG_BC_TESTE_PRINTER

On 3/17/06, Bruno Gomes Pessanha <bruno.pessanha@gmail.com>
wrote:
> I'm migrating a 1.000 queues windows print server with AD very critical
> enviroment to linux/cups/samba solution. Initially, everything working
fine, but
> when reached paroximately 400 queues created the server is getting
degradated.
> The browsing of printers shares is very low. Many times during the day smbd
and
> winbindd need to be restarted to stop the degradation. I didn't found
any
> bottleneck, memory, processor, disk i/o are fine. I have read a lot of
tuning
> docs but nothing solved my problem. Some body knows where can I find more
> information? Or if there's any study case with 1000 queues in a
enviroment with
> 10.000 workstations and aproximately 20.000 users registered in MS-AD. Is
samba
> prepared to this kind of enviroment?
According to this email, CUPS 1.1.x uses a linear lookup for printers,
so it may just be too inefficient to handle 1000 print queues:
http://marc.theaimsgroup.com/?l=samba&m=113165829420082&w=2
CUPS 1.2, which is supposed to fix this, is unfortunately still in beta.

Another print server, such as LPRng, might be able to handle this
better; I don't know.

Josh Kelley