[Samba] Lost performance between Samba 3.0.24 and 3.5.8 with high number of concurrent connections

juan david jd.alarcon at gmail.com
Mon Jun 13 05:20:10 MDT 2011


Hi,

We are trying upgrade our roaming profile sever from Debian Etch to Debian
Squeeze. That's means a upgrade from Samba 3.0.24 to Samba 3.5.8. Our
production environment has above 600 concurrent users without problem. After
upgrade to Samba 3.5.8, server can't manage above 200 users. With
'smbclient', the output is:

Error [user] session setup failed: Call timed out: server did not respond
after 20000 milliseconds

We use Samba+Winbind+kerberos to validate users. After upgrade I have been
tried all configurations that I could imagine without luck.

After that I have tried simplify the problem, I made a test environment with
the next smb.conf:

[global]
   netbios name = yela
   security = share
   guest account = nobody

[mdrive]
   path = /home/HUGU-Profiles/WinXP/enfgen.man
   browseable = yes
   public = yes
   guest ok = yes

Trivial, isn't it? The server is a Debian Squeeze with Linux Kernel 2.6.32-5
and Samba 3.5.8. From one client I ran the next script:

 #!/bin/bash

connectAndList ()
{
        for i in $( seq 1 1000 )
        do
                fechaInicio=$( date )
                salida=$( smbclient //yela/mdrive -U nobody% -c ls 2>
/dev/null )
                retorno=$?
                fechaFin=$( date )
                if [ ! $retorno -eq 0 ]
                then
                        echo "$fechaInicio - Error $1 $salida - $fechaFin"
                        sleep 1
                fi
        done
}

for j in $( seq 1 5 )
do
        for i in $( seq 1 100 )
        do
                 connectAndList $i $j &
        done
done

To summarize, this script do 500 concurrent connections, list the directory
and repeat it. If one connection fail then sleep 1 second and do again.

Samba 3.0.24 run script with 1 or 2 fail connection per second. In Samba
3.5.8 We need falling to 200 concurrent connections, in other case, the
server was freeze and load raise over 20. There isn't any error in log. The
server doesn't fail is the client which has return a timeout.

Maybe the problem are in connection reply, because once you has been connect
with 'smbclient' follow commands work without problem.

Has Some one "Samba" with above 300 concurrent users in production
enviroment? Does some one know something about this performance lost in
connection time?


More information about the samba mailing list