Mailing List Archive

On Tue, Jan 20, 2004 at 06:37:26PM -0500, Jeff Wheeler wrote:
> I've noticed that when changing a neighbor's export policy from any
> configuration, to a configuration that sets metric minimum-igp <offset>
> on an exported route, the advertised MED will not be updated if the
> value would increase as a result, even if the previous configuration did
> not use a minimum-igp based MED value.
>
> Is this the intended functionality, or an implementation quirk?

This is an intended effect of the "minimum-igp" command. As far as I know
it doesn't actually state a reason for this in the documentation, but I
would assume it is to prevent unnecessary route flapping in the event of
an IGP cost change, by not forcing a route update if the metric value
increases past the previous lowest known value.

Of course this begs the question, exactly how long will it stay this way
(forever?) and what happens if you make a permanent routing change which
takes away the lowest cost route? Would you have to clear it in order for
the updates to be processed, and if so would a soft clear out do the
trick? Also, in the event of a sustained outage of the normal lower cost
path (or anything more than a brief flap I would think), it might make
more sense to propagate the new MEDs so that anyone listening to them can
shift traffic accordingly. Perhaps it would make even more sense for there
to be a time limit, where if the cost does not return to its previous
value within X amount of time, an update is sent? Or am I completely
reading this feature wrong?

At any rate, if you don't want the minimum igp cost, leave off the
"minimum-" part. :)

--
Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)

On Tue, 2004-01-20 at 21:39, Richard A Steenbergen wrote:
> Of course this begs the question, exactly how long will it stay this way
> (forever?) and what happens if you make a permanent routing change which
> takes away the lowest cost route? Would you have to clear it in order for
> the updates to be processed, and if so would a soft clear out do the
> trick? Also, in the event of a sustained outage of the normal lower cost

This is what I was trying to indicate in my original post. A soft clear
won't do. A session flap is necessary to effect the change. While your
time limit notion would be useful, perhaps it would be less involved to
implement a command to simply "forget" all the minimum-igp values for a
particular route and re-evaluate its advertised MED for a session or
sessions, or all routes on a specified session. As another example, if
you were to increase the med advertised to a peer from, minimum-igp 50
to minimum-igp 500, that change would not take effect until the routes
were withdrawn from BGP and the local rpd "lost" their state, or until
the affected session flapped / was fully cleared.

> At any rate, if you don't want the minimum igp cost, leave off the
> "minimum-" part. :)

Indeed. :-)

--
Jeff at Reflected Networks

Jeff,

On 01:33 Wed 21 Jan , Jeff Wheeler wrote:
> On Tue, 2004-01-20 at 21:39, Richard A Steenbergen wrote:
> > Of course this begs the question, exactly how long will it stay this way
> > (forever?) and what happens if you make a permanent routing change which
> > takes away the lowest cost route? Would you have to clear it in order for
> > the updates to be processed, and if so would a soft clear out do the
> > trick? Also, in the event of a sustained outage of the normal lower cost
>
> This is what I was trying to indicate in my original post. A soft clear
> won't do. A session flap is necessary to effect the change. While your
> time limit notion would be useful, perhaps it would be less involved to
> implement a command to simply "forget" all the minimum-igp values for a
> particular route and re-evaluate its advertised MED for a session or
> sessions, or all routes on a specified session. As another example, if
> you were to increase the med advertised to a peer from, minimum-igp 50
> to minimum-igp 500, that change would not take effect until the routes
> were withdrawn from BGP and the local rpd "lost" their state, or until
> the affected session flapped / was fully cleared.

in these types of scenario's you could consider to set the metric to igp -> commit -> minimum-igp.

Hth,
-Daniel

>
> > At any rate, if you don't want the minimum igp cost, leave off the
> > "minimum-" part. :)
>
> Indeed. :-)
>
> --
> Jeff at Reflected Networks
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp