Re: HUGE packet-drop

[Rick Jones <rick.jones2 () hp com>]

> The best way I found to do this was to have the application that is receiving
> the packets running on the same cores that the kernel is pulling them off the
> nic.  Since my machine had two chips each with 4-6 cores (8-12 logical
> cores), I limited my application to run on the same chip as the nic was
> receiving its interrupts.

Interesting - because at least for a time there, before LRO/GRO etc, when it was 
near impossible to achieve link-rate with full-size frames on 10GbE, I found 
that in netperf testing I could get higher thoughtput when netserver was bound 
to a core *other* than the one taking interrupts from the NIC - preferably 
though one sharing the last level cache.  I guess in the case of "plain" 
networking (with full-sized segments) the number of cache lines ping-ponged from 
one cache to another is less than the packet capture case.

Rereading the rest though I see you weren't binding the app explicitly to the 
same core, but simply to the same processor, so perhaps the same thing is 
happening with your app as with my netserver :)

rick jones

PS - don't forget that some NICs have multiple IRQs... :)
-
This is the tcpdump-workers list.
Visit https://cod.sandelman.ca/ to unsubscribe.