[37700] in North American Network Operators' Group
Re: Traffic per Prefix Length
daemon@ATHENA.MIT.EDU (Simon Leinen)
Tue May 22 20:34:31 2001
To: Bill Woodcock <woody@zocalo.net>
Cc: "Paul G. Donner" <pdonner@cisco.com>, nanog@merit.edu,
Randy Bush <randy@psg.com>
From: Simon Leinen <simon@limmat.switch.ch>
In-Reply-To: <Pine.GSO.4.21.0105050804290.6533-100000@secure.zocalo.net>
Date: 23 May 2001 02:31:05 +0200
Message-ID: <aaitisc3ee.fsf@limmat.switch.ch>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Errors-To: owner-nanog-outgoing@merit.edu
>>>>> "bw" == Bill Woodcock <woody@zocalo.net> writes:
>> does anyone collect stats that would facilitate the derivation of
>> the number of packets forwarded per prefix length (per unit time)
>> in the routing table or per prefix length range?
> Anybody who's doing both Netflow and BGP table dumps can do that.
> Of course, the corellation between the two isn't perfect, since you
> may have prefix-length changes between snapshots.
Another possibility would be to use one of the bucket-based accounting
mechanisms (BGP Policy Accounting on Cisco and Riverstone, Destination
Class Usage on Juniper):
* define a route-map which maps announcements with different prefix
lengths into different buckets
* have the interfaces count traffic towards prefixes in each bucket
automagically.
The current Cisco implementation only has 8 buckets, which is a bit
low because there are more than 8 prefix lengths in the real world.
Juniper has 16 buckets which looks much better, and Riverstone has 25
which would cover even the absurd cases.
Randy: You could also use this mechanism to count traffic that would
potentially be affected by various kinds of route filters (by
assigning a bucket instead of actually dropping the announcements).
Some pointers to these features can be found on
http://www.switch.ch/misc/leinen/snmp/monitoring/bucket-accounting.html
--
Simon.
