[74222] in North American Network Operators' Group
Re: FW: BGP Load Sharing
daemon@ATHENA.MIT.EDU (Chris Strandt)
Sat Sep 18 09:58:15 2004
Date: Sat, 18 Sep 2004 09:19:45 -0400
From: Chris Strandt <strandtc@liquidweb.com>
To: mmcspedon@arrow.com
Cc: nanog@merit.edu
In-Reply-To: <003701c49d5e$3ebe6bd0$e9fca8c0@LA0721>
Errors-To: owner-nanog-outgoing@merit.edu
This is effectivly what we are doing today, on a slightly more
sofisticated level.
We produce a report based on netflow statistics that tells us (in
percentage terms) which asn's we are sending traffic to. We then look
at our current BGP table, which we get from SNMP. We take the asn's
which recieve 60% of our traffic.. and make a few descisions.
1) If as path length is shorter via one provider over the other, make
sure we prefer the shorter path.
2) If they are equal length, split the list of asn's between the 2
providers based on the input of the script.
So back to the question at hand... to get netflow stats for outgoing
traffic.. we need cards in the 12K router which will support netflow on
the ingress ports of the router for outgoing traffic(ie Gigabit Ethernet
Line Cards)... right?
-Chris
Mike McSpedon wrote:
>>>>What you advertise to upstreams affects incoming traffic, not outgoing.
>>>>
>>>>
>What upstreams advertise to you and the policies you employ on those routes
>affects outgoing. One basic technique is to decide roughly how much outgoing
>traffic you want going over each link, then look at how your current
>outgoing traffic is split (e.g., 70% via ISPA, 30% via ISPB) then manipulate
>the Local Preference attribute on some of the routes you're receiving to
>achieve roughly the balance you're looking for.
>
>
>Mike McSpedon
>Arrow Global Data Communications
>Arrow Electronics, Inc.
>50 Marcus Drive
>Melville, New York 11747
>Phone: 631.847.5551
>
>
>
