[nsp] Counting bps...

[Volodymyr Yakovenko]

On Tue, Mar 18, 2003 at 03:20:02PM +0100, Iva Cabric wrote:
[..skipped..]
>> So whatever you do with whichever counters you use - don't use them for
>> anything that brings money without sanity checks.
>
>Last week, people from Cisco were trying to convince me, that Catalyst
>3550 is definitely the way to go for limiting and accounting (at least
>on Fastethernet).

If they are ready to pay off your loses caused by software bugs - trust them.

Dearly loved telecom manager's principle - always buy complex solution, in
that case only vendor is guilty in all future problems.

>> According to Cisco's documentation on conventional routing architectures
>> byte and packet counters are updated on per packet basis.
>
>Ok, but under heavier CPU loads, are byte and packet counters consistent
>with real values (because of process priorities), or they have some lag?

According to available documentation on traditional architectures (CPU-driven
switching) packet and byte counters are updated during interrupt processing 
by interrupt handler (which is uninterruptable). On shared memory architectures
SNMP counters most likely to be directly mapped by pointer or reference to 
interface counters (IOS does not provide any memory protection between 
processes and interrupt handler, so process may easily reference any memory
object).

However on more complex architectures - 7500, GSR, ASIC-based layer3 switches 
counter's processing seems be much more complex.
 
-- 
Regards,
Volodymyr.