On 12 Apr 2003, Perry E. Metzger wrote:

> [I've set the reply-to to tech-misc to redirect discussion there.]

Ok, though I think -net would be good.

> Havard Eidnes <he@netbsd.org> writes:
> > If this does what I think it does (after browsing the code, I think it
> > does), zeroing the packet statistics will break the monotonous
> > increase of counters restriction imposed on SNMP agents

Doh! Yeah, that'll mess things up.

> Certainly. However, the ioctl in question already existed -- this
> merely exposed it in an ordinary utility program, as you later note.
>
> Really, I wish that there was a NetBSD Technical Review Board that
> would give you an answer to questions like this within a few days so
> that you could just quickly close the argument about particular
> features. Often having a speedy final answer is the most important
> requirement for ending acrimonious technical discussion, even if the
> answer is not 100% perfect by some imaginary metric. (The truth is
> that there often isn't a perfect answer so no 100% perfect solution
> exists anyway and any of several solutions are acceptable provided
> someone actually has the power to make a final decision and end
> debate.)
>
> Unfortunately, core does not seem inclined to answer anything quickly
> (if ever), so we're forced to go by our own judgment.

I don't think we need core for this one.

I think we need two sets of ioctls. One that reads (and possibly resets)
the resettable counters, and another that reads un-resetable counters.
i.e. we keep the counter & the last reset point. The first set reads the
counter minus the last reset (possibly adjusting the last reset), and the
other just reads the counter.

> So, let me note that if you want to modify or remove the feature or
> its implementation, I will not be the least bit upset. I'm more
> concerned with finality and consensus than with the feature.
>
> > I know, this is perhaps more a criticism of the implementation of
> > SIOCZIFDATA than this particular snippet of code, though exposing the
> > knob even more (as is done here) makes this more of a problem.

Indeed.

So what do other *BSDs do? I figure we need to come up with a new
read-stats ioctl that will get the monotonic value. What name(s) are in
use now?

Take care,

Bill