On Fri, Oct 15, 1999 at 05:28:15AM +0900, itojun@iijlab.net wrote:
> 
> >>>In the netstat outpur below, the Ibytes and Obytes for both address
> >>>families are the same. Is this a bug, something not yet implemented
> >>>or do I miss some option in my kernel to get it to discriminate
> >>>between address families?
> >>	I believe there's no code for per-AF, per-IF stats.  Ibytes and Obytes
> >>	are purely per-if stats (not per-AF).
> >	I was wrong.  NetBSD has struct ifaddr_data (sys/net/if.h) which does
> >	per-ifaddr stats.  I'll be adding some support for this.
> 
> 	Sorry again, I was right.  per-ifaddr stats are never used
> 	(IFA_STATS never defined).  code fragment in ip_output() will panic
> 	the kernel as it chases dangling pointer.

Is it possible that the per-if netstats count the number of bytes
twice (maybe for both IP and IPv6)? On my machines it looks like 
the ibytes and obytes are a factor 2 too high. My reference is
stats collected from a cisco router. The NetBSD machines and the
cisco seem to agree on the number of packets though.

Feico.