Greg A. Woods <woods@weird.com> wrote on Sun, 13 Apr 2003
at 15:59:34 -0400 in <m194ndW-000B3kC@proven.weird.com>:

> > it's not as convenient as zeroing the counters. You can come 2 hours, or a
> > day after zeroing them and see how they moved.
> 
> Well if you're going to wait that long, and if you're going to sample by
> eye at random times, then you really want to have some way to remember
> when exactly it was you started counting, no?

Probably not. Generally you'll clear counters at a particular
event, like "We just finished rewiring the network and all the errors
we saw prior to this time are no longer of interest because they were
because of this bad cable we replaced."

And then you'll want to wait some period of time after that to check
and see if things look plausible. And then you'll care about the ratio
of errors to input bytes (or somesuch), and the exact time is not
necessarily of relevence.

At least, that's been my experience with the clearing of counters.
But yes, knowing the time they were cleared at is of use too.

But the overriding point is that it's not always easy, from a user
interface perspective, to run a persistent process once you start
clearing the counters, and then keep coming back and looking at it.
Users would prefer to be able to run the process once, at the time of
inquiry, and then be able to logout and pursue their business, but
then go back and run the inquiry command again. All without having to
put extra effort into saving the values of the counters (in their
head, on a file, on paper, etc.) at the initial time of clearing.


So, while I don't strongly support the idea of clearing counters,
I think we know why people want to do it, and if we do implement it,
we should implement it such that the persistent counters are available,
and the time of counter-clearing is also available.

--jhawk