samba slowdown

[tunjin]

I have some serious speed-problems using sama (3) on a suse 9.0 installation.
When i copy files from a windows xp machine, the suse machine shows ~20-40kb total speed in iptraf and the copy process lasts about 30min for a 250 Meg file...
All the usual registry settings are applied on the xp machine and the TCP_NOWDELAY... are set on the smb server.

The odd thing: everytime i try to copy a file over smb, the number of dropped packages increases:

eth0      Link encap:Ethernet  HWaddr 00:40:63:D3:8C:C5
          inet addr:192.168.0.100  Bcast:192.168.0.255  Mask:255.255.255.0
          inet6 addr: fe80::240:63ff:fed3:8cc5/64 Scope:Link
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:318506 errors:0 dropped:54088 overruns:0 frame:0
          TX packets:296966 errors:2 dropped:0 overruns:2 carrier:0
          collisions:0 txqueuelen:100
          RX bytes:401877556 (383.2 Mb)  TX bytes:48714868 (46.4 Mb)
          Interrupt:10 Base address:0xe800

when i run slackware on the same machine as xp, i get good speed and no dropped packages. copying the file from the xp machine over sftp also results in good speed....
currently there is no firewall running on the suse machnie..

any help / pointers?

[wassy121]

Let us see the output of mii-tool . Also, are these two switched or hubbed together? I think it might be due to the card auto-negotiating to 10 Mb HD or FD, instead of 100 FD.

Best,

Samuel

[tunjin]

output - as expected both 100mbit - also even a 10mbit hubbed solution would not explain, why smb is dog-slow (40kb, only on windows-xp) and sftp is at normal speed.

linux:~ # mii-tool
eth0: negotiated 100baseTx-FD flow-control, link ok
eth1: negotiated 100baseTx-FD, link ok

thx for your suggestion though

--edit--
additional info: eth0 is the internal nic, eth1 is connected to my dsl modem.
samba is set up to work only on eth0 (though, the problem existed before i configured that..)
--edit--

[tunjin]

additional information:
the problem exists only if i write to the samba server - and is the same for windows / linux clients.
the strange thing is, as mentioned before, that a huge amount of packets is dropped by the kernel, though there is no firewall running. the normal cause would be a slow cpu, but load is ~0.15 on this machine and smbd takes at most ~4% cpu...
hardware: via mini-itx board, eden 533 cpu, 256mb ram, hdd's with 8meg cache, eth0 is the onboard nic...