[45089] in North American Network Operators' Group
RE: router startup behavior
daemon@ATHENA.MIT.EDU (Borchers, Mark)
Wed Jan 16 09:49:25 2002
Message-ID: <CA47B6D616C0D211B92E0008C7C5657C16F5FF61@hscmpxsrvcl01>
From: "Borchers, Mark" <mborchers@splitrock.net>
To: 'Havard Eidnes' <he@nordu.net>
Cc: ratul@cs.washington.edu, nanog@merit.edu
Date: Wed, 16 Jan 2002 08:48:51 -0600
MIME-Version: 1.0
Content-Type: text/plain;
charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
Errors-To: owner-nanog-outgoing@merit.edu
> > Just guessing - you're seeing these events between midnight=20
> > and 5 am?
>=20
> Hm, couldn't reist this one: "which time zone"?
>=20
> Just hinting that even though it's that time interval in the US, =
local
> time is different in other places around the world, so if this is
> causing disturbance, others are probably being hit in their working
> hours.
Heh, the "perpetual global maintenance window syndrome", eh? A
very useful concept...
=20
> Besides, I was under the impression that to activate a new outbound
> roting policy on a Cisco, you could just modify / replace it, but =
that
> you would still have to do
>=20
> router#clear ip bgp xxx soft out
>=20
> to activate it. This means that the policy for an existing peer can
> be modified without having to remove the peering and reenable it
> shortly thereafter (something which would cause needless route
> flapping).
Somewhat true. The new policy would not be applied to routes that=20
were already in the table, but would be applied to any adds/withdrawals
that occur once the policy changes are placed in the configuration.
This fact has a synergistic effect when you're making changes that
affect lots of sessions. So a policy change made a significant time
before a clear or soft clear could in fact result in flaps.
Also consider: there are a lot of routers using traditional ACL's in =
their=20
policies (as opposed to things like prefix-lists which have more =
granular
editing features) which would necessitate removing the ACL completely =
and=20
rewriting it with updated lines. Due to the above, a potential for =
leaks=20
exists unless the session is either shutdown or deleted while acl's are =
being modified.=20
If the sequence of events in a configuration script is not well thought
out, the result could be what Ratul has observed in his study.=20
>=20
> Regards,
>=20
> - H=E5vard
>=20
