[Soekris] Strange problem with net4801

Horst Laschinsky horst.laschinsky at physik.uni-erlangen.de
Tue Jan 18 16:42:33 UTC 2005

Jim Shankland wrote:
> (2) Packet size.  When a link or interface card is flakey, often loss
> is worse for larger packets.  As noted above, tftp uses 512-byte
> packets.  Any tcp-based transport will use 1500-byte packets on
> a directly connected Ethernet.

I forgot to mention in my original posting, that I already did some
tests on this issue. I decreased the mtu (of both interfaces) in 
certain steps until I reached the ethernet minimum, but also this
had no influence. The effect remained.
> Try running some ping tests with various different packet sizes:
> I'll bet things get much worse as packet size increases, and that
> one direction is much worse than the other.

Sorry, you're wrong. Even flood ping with packet size of 1492
(+ 8 byte icmp header -> 1500 byte ethernet mtu) shows no effect:

Ping from Soekris to the PC:

soekris:~# ping -f -s 1492
PING ( 1492 data bytes
--- ping statistics ---
6949 packets transmitted, 6938 packets received, 0% packet loss
round-trip min/avg/max = 0.0/1.1/10.0 ms

Ping from PC to the Soekris:

linux:~ # ping -f -s 1492
PING ( 1492(1520) bytes of data.
--- ping statistics ---
20409 packets transmitted, 20385 received, 0% packet loss, time 11900ms
rtt min/avg/max/mdev = 0.452/0.536/9.805/0.165 ms, ipg/ewma 0.583/0.539 ms
> Why?  Barring some kind of driver or cable problem, it's flakey
> hardware.

I still have not the faintest clue what could be wrong. Seems that
only ssh/scp causes this effect - but again: how can it influence
the physical layer?

Best regards,


                             Horst  Laschinsky
        Universitaet Erlangen-Nuernberg - Physikalisches Institut I
 Erwin-Rommel-Str. 1, Room: 207, 91058 Erlangen - Tel: ++49 9131 85-27076

More information about the Soekris-tech mailing list