Cold start ethernet problems

Please also dump the result of ethtool -S eth0 when error happens.

Check if any overflow is seen in the statistics.

Just tested it, it reports overflow eth_problem_dump.txt (5.2 KB)

Please try this setting and see if it can enhance.

echo 2140004608 > /proc/sys/net/ipv4/ipfrag_high_thresh
sysctl -w net.core.rmem_max=26214400
sysctl net.core.rmem_max
sysctl -w net.core.netdev_max_backlog=2000
sysctl -w net.core.netdev_budget=600

On the problematic unit, these had no effect.

Another development is that i witnessed the same problem on the other Xavier unit after running the program for ~30 minutes. On this unit the problem did not get big enough to crash the program, but many images were skipped.
While the problem persisted, i witnessed that ksoftirqd/0 was using ~30% of the cpu core.
After trying the settings on this unit, at the same time enabling jetson clocks, the problem seems to have disappeared.

If this happens on different xavier hw with same sw configuration, I would like to suggest you to RMA this problematic board.

Ok, i will also try to reproduce the problem on a third Xavier unit we have, which has jetpack 4.4 installed.

Hi again,
I have now tested with multiple Xavier units, including one with jetpack 4.4, And on all of them the problem occurs with varying degrees of intensity.

We also bought a PCI-E ethernet card, a relative cheap on at that ( tp-link TG3468) which is supported on the Xavier. With this card i experience no issues at all. It is a bit unpractical for us to use this PCI-E card for our purposes with the dev kit. Is there any other alternatives to configure the standard ethernet adapter?