[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Call for censensus on path forward

On Sat, 21 Sep 2002 14:00:42 +0200, "Wijnen, Bert (Bert)" <bwijnen@lucent.com> said:
> I would really like to hear from a few operators (or
> even NM application developers) if they indeed find it
> a requirement to do polling a 1-second-granularty.
> It does not sound realistic to me... but who is me?

As I said, I sometimes use a homewritten tool that polls all
if(HC)In/OutOctets from a router every few seconds and prints a table
of the traffic rates in full-screen mode using curses.

This is handy when I have to re-route traffic for some kind of
maintenance, and want to verify that my IGP metric changes had the
desired effect.

If my routers updated their interface counters as often as every
second, then I would probably use 2-5 seconds as the polling interval
for this tool.  Right now I have to use 10 or even 20 seconds with
some routers, because otherwise the rates will flip-flop between zero
(or almost zero) and multiples of actual loads.  There is probably
some kind of an implementation/performance reason why the SNMP
counters aren't updated more often from the (hardware) interface
counters, and that might not go away by adding clever mechanisms.

For what it's worth, on my old Linux laptop with Wes' SNMP agent I can
poll the counters ten times a second, and the rates always look
realistic.  But then it doesn't really have as many or as fast
interfaces as our routers.