[isp-bgp] Re: MEDs from different ASes

[Iljitsch van Beijnum <iljitsch@...>]

On 20-mei-04, at 15:40, Tulip Rasputin wrote:

> I am aware of how we can configure the boxes so that the MEDs are 
> always
> compared, even across the ASes. What i wanted to know was, *why* would
> somebody ever want to do that?
Didn't we have this discussion a few of weeks ago?

It's very common today to see the same AS path length for 90% or so of 
all destinations from two or more ISPs. In the absense of any 
configuration that tells the router otherwise, the selection of the 
outbound ISP then boils down to the four tie breakers listed in RFC 
1771, which often results in way too much traffic flowing over one ISP.

This is fixable by overruling the "lower" tie breakers by manipulating 
the local preference, AS path or some more obscure values such as 
origin or weight. Or the multi exit discriminator, if 
always-compare-med is set.

Now some people argue that the local preference is the appropriate 
value to change in these cases. I can see the logic where all pertinent 
data is put through a formula that has a local preference value as its 
outcome. In this case it would be appropriate to manipulate this value 
to balance the traffic. However, AFAIK only Bay routers work this way. 
Other BGP implementations leave the local pref alone by default. So if 
you change the local pref for some routes, you're overruling all other 
path selection criteria with this. IMO, this is not appropriate when 
balancing traffic over more than one ISP as you're basically hardcoding 
routing information.

Then there is the AS path. This works well if you're a leaf network, 
but if you provide transit service, it may annoy your customers as it 
messes up the AS paths they see. Also, you may still be in the 
situation that many AS paths over different ISPs are the same length so 
a change in the remote router ID or the peer adress can have a big and 
unexpected impact on the traffic flow.

So this leaves the MED as a convenient value to change in order to 
balance outgoing traffic, as it doesn't overrule the AS path and the 
new MED is usually invisible to customers.







To unsubscribe via postal mail, please contact us at:
Jupitermedia Corp.
Attn: Discussion List Management
475 Park Avenue South
New York, NY 10016

Please include the email address which you have been contacted with.