[quagga-users 889] Re: AW: AW: getting quagga

Paul Jakma paul at clubi.ie
Fri Nov 14 03:00:31 GMT 2003

On Thu, 13 Nov 2003, Rohit Gupta wrote:

> > 
> >         AS1
> >        /   \
> >       AS2  AS3
> >       | \  / |
> >       |  \/  |
> >       \  /\ / 
> >        R1  R2
> > 

> I dont find not having MEDs as transitive a problem at all. The
> example which you quoted explains why.

Possibly. :)

> Imagine what would have happened if the MEDs were transitive. You
> would be deciding upon the routes based on AS1's internal topology
> regardless of how AS2 and AS3 can reach those routes.

No, the MED's do not express AS1s internal topology, they express 
AS1s preferences for traffic weighting on its links to its peers.

> Thus MEDs can and should only be used when recieving them from the
> originator and should never be compared across the ASes 


> (as you so aptly explained).

> And If i didnt stress enough, MEDs have no use if they are made
> transitive.

Well thats not quite true. Imagine if MEDs were specified as follows:

- can only be set by originating AS (noone else may modify)
- are transitive

Say AS1 sets MED of 10 on announcements to AS2, and 100 on AS3. Then:

R1 gets:

	AS2_AS1 10
	AS3_AS1 100

R2 gets:

	AS2_AS1 10
	AS3_AS1 100

And suddenly the MED becomes a lot more useful :) (least in this
case) - if this were the case, MED's /could/ be compared across ASes.
In the above case, it would mean that we (and all other hosts who
find AS1 is equal as path length distance away) will choose the path
which is closer to AS1's preferred inbound link, ie via AS2. Which
should mean AS1 gets more traffic via AS2. This is I suspect how a
lot of people think MED works.

Course, the MED is no longer useful for 'us' to do path weighting on
traffic from AS2 to 'us' (or from AS3). So really we need 2 kinds of
MED, one transitive, to allow a preference to be propogated globally
(with 'originator only can set' restrictions as above), eg MED_GLOBAL
and one non-transitive, ie the existing MED.

Effects of such a MED_GLOBAL though can be achieved with more control
via AS path prepending, which is i guess why it doesnt exist. :)

> Regards,
> Rohit

Paul Jakma	paul at clubi.ie	paul at jakma.org	Key ID: 64A2FF6A
	warning: do not ever send email to spam at dishone.st
Real Users are afraid they'll break the machine -- but they're never
afraid to break your face.

More information about the Quagga-users mailing list