Re: Interface counter inaccurate

On 16 Feb 2021, at 0:58, Daniel Ponte wrote:
> On Mon, Feb 15, 2021 at 10:25:47PM +0100, Kristof Provost wrote:
>> On 15 Feb 2021, at 22:09, Daniel Ponte wrote:
>>> I've noticed that since upgrading to stable/13-n244514-18097ee2fb7c 
>>> from
>>> 12.2-STABLE, throughput on my WAN interface (the box runs pf) is
>>> incorrectly showing double in systat -if, as well as in vnstat and 
>>> pftop
>>> from ports. The LAN interface does not appear to be so afflicted.
>>>
>> `systat -if` doesn’t read the pf counters, so I wouldn’t expect 
>> that to be
>> related.
>>
>> Those are the interface counters.
>> What network card and driver do you use?
>>
>> Kristof
>
> I, too, questioned the relation. They are igb(4) I210 builtin 
> interfaces
> (in a Protectli Vault 4).
>
>
> systat -if during said speed test:
>
>            igb1  in     11.670 Mb/s         11.670 Mb/s            
> 1.067 GB
>                  out   319.975 Mb/s        320.458 Mb/s            
> 2.062 GB
>
>            igb0  in    640.120 Mb/s        640.690 Mb/s            
> 6.351 GB
>                  out     5.824 Mb/s          5.824 Mb/s          
> 987.050 MB
>
> igb1 is inside, igb0 is outside. The 6GB:2GB difference in totals seen 
> above
> is indeed real; this machine did not initiate that much traffic on its 
> own.
>
Even stranger, in that you appear to only have the issue on one of your 
igb0 interfaces.
My initial guess was that there was a driver bug causing it to double 
count the packets.

That machine has 4 ethernet ports, does it have 4 igbX interfaces as 
well? Are there any vlans configured? (My current thinking is still that 
it’s a driver issue, manifesting only on one of the interfaces because 
of a configuration difference.)

It may also be useful to try capturing packets on igb0 and correlating 
the number of captured packets with the counters.

I’m not all that familiar with the igb driver code, so I don’t know 
if I’ll be able to help much.

Best regards,
Kristof