Eth0 (1GB, no traffic) and heavy IO traffic cause hang/reboot

Hi,

this is a (frustrated) report of the guruplug server causing a hang (or seldom
reboot) when heavy IO traffic is combined with an 1GB connection (no traffic) on
a single eth port. My guruplug server plus is running debian squezze, controlled
by a JTAG console connection at 230V AC. The box is brand new ~1 month, running
only 20-30 hours.

I recognized that the box was hanging (power consumption get down by 1.5-2.5 W,
no console communication is possible any more, power off necessary to restart)
at certain tasks. After some experiments, I was able to "reliable" hang the box
after about 10-60 sec of traffic, independently of the current box
temperature. BTW, i measured 56C (degree celsius) @ 5.2 W, and 61C @ 6.5W, near
the u-snap connector with an infra-red thermometer.

The hang can be reproduced with either a transcend 8GB class 6 micro SDHC card or
an external USB2 hard drive (externally powered). Perhaps others can reproduce
with even less equipment.

My experiments combined 3 networks setups with 3 IO-traffic setups:

Network setups:
n0: running with no ethernet cable
n1: 1x ethernet 100M connected, no traffic (IP assigned or not doesn't matter)
n2: 1x ethernet 1GB connected, no traffic (IP assigned or not doesn't matter)

IO setups:

io0: idle, no IO traffic
io1: nand to SDHC (dd if=/dev/ubi0_0 of=/media/usb1/junk bs=1M count=500)
io2: nand to ext USB (dd if=/dev/ubi0_0 of=/media/usb2/junk bs=1M count=500)
io3: SDHC to ext USB (dd if=/dev/sdb of=/media/usb2/junk bs=1M count=500)

Here is what i observed:

- n0: No eth cable at all
io0: idle: 5.0 W <OK>
io1: nand to SDHC:    5.7-5.9 W <OK>
io2: nand to ext USB: 5.9-6.0 W <OK>
io3: SDHC to ext USB: 5.6-5.7 W <OK>

- n1: 1x eth 100M, no traffic
io0: idle: 5.2 W <OK>
io1: nand to SDHC:    6.0-6.2 W <OK>
io2: nand to ext USB: 6.1-6.2 W <OK>
io3: SDHC to ext USB: 5.9 W <OK>

- n2: 1x eth 1G, no traffic
io0: idle:      6.5 W <OK>
io1: nand to SDHC:    7.2-7.4 W <*** reboot after 8 sec, hang after 30, hang after 65 sec ***>
io2: nand to ext USB: 7.5 W <*** hang after ~35 sec ***>
io3: SDHC to ext USB: 7.1 W <*** hang after ~30 sec ***>

The net observation is, that as soon as you have a 1GB ethernet connection,
additional IO traffic will hang (or reboot) your guruplug.
For me, this renders the ethernet 1GB unusable, i don't even tried with 2 of
them connected.

I would like to hear what others do see. Perhaps one can write some nasty
IO-traffic script just using the nand to hang the machine.

This guruplug sucks. Any news about a firmware fix (if
possible) would be nice, but i doubt.

Greetings

Christoph