> just a note that most normal devices do not use more then 32 bits for
> traffic counters, and I include the high-end Cisco routers in this. They
> figure that if you are monitoring the traffic numbers and know when they
> will wrap (at 32 bits) you can detect the wrap and include it in your
> stats, they even make comments about the need to monitor the really high
> speed ATM connections more frequently then others so that you can detect
> the wrap anc account for it.
This is a bad assumption that people make... as interface speeds increase,
counter wrap occurs much more frequently. Just because I have an OC-12
interface, doesn't mean I should have to poll it ever 45s to get an accurate
coutner reading.
I've been hounding Cisco, etc to use either 64-bit counters, or keep an
internal 8 or 16-bit (or 32-bit) "wrap counter" for their interfaces for
ages. At least routers will report things like "avg bps over last 5-min
interval".
ANyway, interface speeds will only increase, so there's nothing good about
saying "just poll more often" -- pretty soon all your traffic will be polling
traffic ;-)
Better to fix the problem (not enough bits in the counter) than to try to
work around it. The 'wrap counter' isn't all that bad, but it just
complicates things (requires 2 polls, instead of 1)
-Taner
--
D. Taner Halicioglu taner@globalcenter.net
Network Engineer Frontier GlobalCenter
Tel: +1 (408) 328-4729 1154 East Arques Ave
Fax: +1 (408) 328-4806 Sunnyvale, CA, 94086
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html