> grep "reassembles failed" /tmp/1 /tmp/2 /tmp/3 /tmp/4;
> /tmp/1: 608545 packet reassembles failed
> /tmp/2: 881139 packet reassembles failed
> /tmp/3: 1128584 packet reassembles failed
> /tmp/4: 1322635 packet reassembles failed
Very interesting result. The dependence is inverted. 8)8)
Probably fragments are dropped due to backlog overflows.
The more reassemblies succeed is the more time we lose merging,
the more pressure on backlog. Negative feedback, equilibrium value
is unpredicatble. 8) Look into /proc/net/softnet_stat,
the second column. What does it show?
> track it down. When I used DaveM's zero-copy patches on
Wow! Did you test _without_ it?
How was zerocopyless kernel able to hold 1.5Gig? It copies all twice!
> a linux-2.4.0 kernel, most, if not all, of these
> packet receive errors went away.
They must disappear completely. Each loss on LAN is bug.
User has right to expect that no losses happen.