P.Elford@aarnet.edu.au (Peter Elford) (04/20/91)
The Australian Academic and Research Network (AARNet) is currently
configured as a star of stars, ie.
QLD
|
| NSW
ACT | /
\ | /
\|/
WA -----------N---TAS
/| P----> 128K satellite to USA
/ | |
/ -+---+---+-
SA |
VIC
Each of the acronyms represents an AGS router that has links to all
the AARNet members within a particular state. The "N" is also an AGS
(the national hub) and "P" is a Proteon that terminates our link to
FIX-WEST. All but one of the routers is runing 8.1(19) (VIC is an AGS+
with 8.2(1)). All the links are 48K except for a short ethernet between
N, VIC and P, and the link to the US.
Routing in this enviroment is VERY simple. We use RIP to pass network
routes up from the individual member sites to the state hubs, and from
them to the national hub. The national hub, VIC and the proteon are the
only routers that have a route for all Australian network numbers; all
the other state hubs only have routes to the networks within their state.
The router at each member site has a static route to default that points
at the state hub and the state hubs have a static route to default that
points at the national hub. The Proteon sends out a route to 132.160
(the PACCOM network that links us to the US) and the national hub uses
this as its default network.
We are about to install two E1 circuits as shown below:
ACT NATIONAL
+---------------+ 2Mb +---------------+
NSW | 139.130.188.2 +---------+ 139.130.188.1 |
+---------------+ 2Mb | | | |
| 139.130.192.2 +---------+ 139.130.192.1 | | |
| | 48K +---------------+ | |
| 139.130.72.2 +-----------------------------------+ 139.130.72.1 |
+---------------+ +---------------+
The 48K between ACT and the national hub will be retired when the 2Mb lines
go in, but the 48K between NSW and the national hub will be retained for
backup.
HOW DO WE SET UP OUR ROUTING TO DO DYNAMIC FAILOVER TO THE 48K IF ONE
OF THE 2Mb LINKS DIES ?
My current thinking goes as follows:
A) Stick with RIP. We can only do this if there is some method of
artificially increasing the hop count of routes sent from NSW to
the national hub over the 48K line. The "offset-list" parameter only
allows you to fiddle routing metrics on a per network, rather than
per interface, basis.
IS THIS TRUE ? CAN WHAT I WANT BE DONE ?
B) Fire up IGRP to be able to dynamically detect the failure of one of
the 2Mb lines. NSW and ACT routes will have to distributed between all
three routers but since most of them have are learnt via RIP, to
redistribute them using IGRP we will have to use a "default-metric"
for them all.
DOES THIS MEAN THAT ALL RIP ROUTES REDISTRIBUTED BY IGRP WILL HAVE THE
SAME METRIC REGARDLESS OF WHICH INTERFACE THEY GO OUT OVER ?
(which, of course, negates the reason why we were going to run RIP in
the first place :-).
I guess the another option would be to run IGRP throughout AARNet, but
that's a pretty big change just for a new link!
Also, we will have to be a little smarter about how "default" is
handled. Obviously the static default routes will have to be removed in
the ACT and NSW routers. Since ALL routers in the network ignore any
RIP default routes sent to them (over any interface, remember we are
running 8.1 ...) we can probably safely use IGRP to pass around a
dynamic default route. It would be nice to use the PACCOM network as
our "default-network" but that would mean that if the satellite link
went down then ACT and NSW would not have a route back to the national
hub. We could of course pass routes for all Australian network numbers
to ACT and NSW from the national hub, but we have just put in place a
whole bunch of filters to avoid this (and the issue of metrics when
redistributing RIP to IGRP appears again, but this is less of a problem
in this case since the other states are all still in a star topology).
ARE THERE ANY OTHER WAYS OF PASSING DEFAULT GATEWAY INFORMATION AROUND ?
Would this work ?
router igrp 1221
redistribute static
default-network 132.160.0.0
!
ip route 132.160.0.0 139.130.x.y
ie. advertise 132.160 via IGRP, marked as an "exterior route" but don't
rely on it being passed dynamically to us from the Proteon.
Finally, on a related issue:
DOES AN IGRP "EXTERIOR ROUTE" OVERRIDE A RIP DEFAULT NETWORK (0.0.0.0) ?
(I assume so because IGRP routes have a smaller administrative distance
than RIP derived ones by default).
Comments gratefully appreciated,
Peter Elford, e-mail: P.Elford@aarnet.edu.au
Network Co-ordinator, phone: +61 6 249 3542
Australian Academic Research Network, fax: +61 6 247 3425
c/o, Computer Services Centre, pager: +61 6 245 3035
Australian National University post: PO Box 4
Canberra, AUSTRALIA Canberra 2601tcs@uunet.UU.NET (Terry Slattery) (04/22/91)
Peter,
Redrawing your network to reflect the new configuration (note the
transposition of QLD and ACT to make the drawing simpler):
2Mpbs
ACT---------+
| |
| NSW
|2Mbps |
QLD | |.048Mbps
\ | ---------+
\|/
WA -----------N---TAS
/| P----> 128K satellite to USA
/ | |
/ -+---+---+-
SA |
VIC
As I understand it, you want to have ACT and NSW use the ACT->N path for
normal operation. ACT should switch to the ACT->NSW->N path if the ACT->N
path fails and NSW should use the NSW->N path if either ACT->N or NSW->ACT
path fails.
>HOW DO WE SET UP OUR ROUTING TO DO DYNAMIC FAILOVER TO THE 48K IF ONE
>OF THE 2Mb LINKS DIES ?
>
> My current thinking goes as follows:
>
>A) Stick with RIP. We can only do this if there is some method of
>artificially increasing the hop count of routes sent from NSW to
>the national hub over the 48K line. The "offset-list" parameter only
>allows you to fiddle routing metrics on a per network, rather than
>per interface, basis.
>
>IS THIS TRUE ? CAN WHAT I WANT BE DONE ?
Your problem is that you want to make routing decisions based on link
bandwidth and the protocol you're using uses hop count as its metric.
Something outside RIP must be done to add the information about path
bandwidth.
You could specify a floating static route to install the backup path with an
administrative distance which is normally overriden by RIP updates from the
routers along the 2Mbps path. If the RIP info is lost, then the path is
presumed down and you use the backup static route. This allows you to stay
with RIP and all of your current configuration.
>B) Fire up IGRP to be able to dynamically detect the failure of one of
>the 2Mb lines. NSW and ACT routes will have to be distributed between all
>three routers but since most of them have are learnt via RIP, to
>redistribute them using IGRP we will have to use a "default-metric"
>for them all.
>
>DOES THIS MEAN THAT ALL RIP ROUTES REDISTRIBUTED BY IGRP WILL HAVE THE
>SAME METRIC REGARDLESS OF WHICH INTERFACE THEY GO OUT OVER ?
Yes, all RIP routes will have the same metric if you redistribute them with
IGRP. This is typically not a problem since the information will disappear
if the route is marked down.
You could do some weird things using different IGRP Autonomous Systems,
selecting which routes each IGRP AS includes, and have each redistribute via
RIP. This is probably much more complex than you want.
>(which, of course, negates the reason why we were going to run RIP in
>the first place :-).
>
>I guess the another option would be to run IGRP throughout AARNet, but
>that's a pretty big change just for a new link!
Why does that negate your reason for running RIP? You're using default
routes for all out-of-area routes, so all you have is reachability
information (no metric).
Most places run IGRP in their routers and redistribute the information using
RIP. The IGRP routers broadcast RIP default routes tell the stub networks of
the backbone. When you're running IGRP, redistributing the IGRP information
into RIP, and need to import some dynamic routes via RIP, you have to
install filters to select only the RIP routes you are interested in
propagating. (This presumes that the equipment sourcing the RIP routes
doesn't do split-horizon.)
>Would this work ?
>
>router igrp 1221
>redistribute static
>default-network 132.160.0.0
>!
>ip route 132.160.0.0 139.130.x.y
>
>ie. advertise 132.160 via IGRP, marked as an "exterior route" but don't
>rely on it being passed dynamically to us from the Proteon.
That looks fine to me.
>Peter Elford, e-mail: P.Elford@aarnet.edu.au
-tcsjyy@merit.edu (Jessica Yu) (04/26/91)
It would be easier for your situation if cisco software allows user to define
default-metric' per interface. Something like:
router igrp 1000
default-metric - - - - - serial 0
default-metric - - - - - serial 1
redistribute rip
...
(cs@cisco.com : could you do this?)
Without that function, one can still do some kludge. The one I think may work
for you is to use two IGRP sessions within the three ciscos (NSW,ACT and N) and
use passive interface subcommand and adminstrative distance to favor one
IGRP session than the other. The idea is to use two IGRP sessions to
distinguish routing information from different path therefore enable the
router to select the favorable path. For example on NSW router in your
picture, one can do the following:
router igrp 1001
default-metric - - - - -
redistribute rip
passive interface s 0 (the one connect to the 48K link)
router igrp 1002
default-metric - - - - -
redistribute rip
passive interface s 1 (the one connect to the 2mb link)
Suppose interface s 0 is the one with 48K link and s 1 is the one with 2mb link.
Router N will learn the same set of routes behind NSW router via both IGRP 1001
and IGRP 1002 and we know the routes learned via IGRP 1001 is sent out via s1
which connects the fast path. By applying adminstrative distance to favor
IGRP 1001 on router N, it will send traffic to the fast path when it is not
down and IGRP 1002 routes as backup. Same kind of thing need to be done on
N to control the routing information (a default from your description) from the
other direction.
Have fun!
--jessica
From root Sat Apr 20 01:31 EDT 1991
Received: Sat, 20 Apr 91 01:31:49 -0400 from merit.edu by kiddo.merit.edu (AIX/1.6)
Return-Path: <cisco-request@spot.Colorado.EDU>
Received: from spot.Colorado.EDU by merit.edu (5.65/1123-1.0)
id AA25676; Sat, 20 Apr 91 01:30:57 -0400
Received: by spot.Colorado.EDU (5.57/CU-CNS2.4-C)
id AA28819; Fri, 19 Apr 91 22:41:44 -0600
Received: by spot.Colorado.EDU (5.57/CU-CNS2.4-C)
id AA28306; Wed, 17 Apr 91 18:54:56 -0600
Received: by jatz.aarnet.edu.au id AA13406
(5.65+/IDA-1.3.5 for cisco@spot.colorado.edu); Thu, 18 Apr 91 10:54:18 +1000
Date: Thu, 18 Apr 91 10:54:18 +1000
From: Peter Elford <P.Elford@aarnet.edu.au>
Message-Id: <9104180054.AA13406@jatz.aarnet.edu.au>
To: cisco@spot.Colorado.EDU, cs@cisco.com
Subject: Routing query
Status: RO
The Australian Academic and Research Network (AARNet) is currently
configured as a star of stars, ie.
QLD
|
| NSW
ACT | /
\ | /
\|/
WA -----------N---TAS
/| P----> 128K satellite to USA
/ | |
/ -+---+---+-
SA |
VIC
Each of the acronyms represents an AGS router that has links to all
the AARNet members within a particular state. The "N" is also an AGS
(the national hub) and "P" is a Proteon that terminates our link to
FIX-WEST. All but one of the routers is runing 8.1(19) (VIC is an AGS+
with 8.2(1)). All the links are 48K except for a short ethernet between
N, VIC and P, and the link to the US.
Routing in this enviroment is VERY simple. We use RIP to pass network
routes up from the individual member sites to the state hubs, and from
them to the national hub. The national hub, VIC and the proteon are the
only routers that have a route for all Australian network numbers; all
the other state hubs only have routes to the networks within their state.
The router at each member site has a static route to default that points
at the state hub and the state hubs have a static route to default that
points at the national hub. The Proteon sends out a route to 132.160
(the PACCOM network that links us to the US) and the national hub uses
this as its default network.
We are about to install two E1 circuits as shown below:
ACT NATIONAL
+---------------+ 2Mb +---------------+
NSW | 139.130.188.2 +---------+ 139.130.188.1 |
+---------------+ 2Mb | | | |
| 139.130.192.2 +---------+ 139.130.192.1 | | |
| | 48K +---------------+ | |
| 139.130.72.2 +-----------------------------------+ 139.130.72.1 |
+---------------+ +---------------+
The 48K between ACT and the national hub will be retired when the 2Mb lines
go in, but the 48K between NSW and the national hub will be retained for
backup.
HOW DO WE SET UP OUR ROUTING TO DO DYNAMIC FAILOVER TO THE 48K IF ONE
OF THE 2Mb LINKS DIES ?
My current thinking goes as follows:
A) Stick with RIP. We can only do this if there is some method of
artificially increasing the hop count of routes sent from NSW to
the national hub over the 48K line. The "offset-list" parameter only
allows you to fiddle routing metrics on a per network, rather than
per interface, basis.
IS THIS TRUE ? CAN WHAT I WANT BE DONE ?
B) Fire up IGRP to be able to dynamically detect the failure of one of
the 2Mb lines. NSW and ACT routes will have to distributed between all
three routers but since most of them have are learnt via RIP, to
redistribute them using IGRP we will have to use a "default-metric"
for them all.
DOES THIS MEAN THAT ALL RIP ROUTES REDISTRIBUTED BY IGRP WILL HAVE THE
SAME METRIC REGARDLESS OF WHICH INTERFACE THEY GO OUT OVER ?
(which, of course, negates the reason why we were going to run RIP in
the first place :-).
I guess the another option would be to run IGRP throughout AARNet, but
that's a pretty big change just for a new link!
Also, we will have to be a little smarter about how "default" is
handled. Obviously the static default routes will have to be removed in
the ACT and NSW routers. Since ALL routers in the network ignore any
RIP default routes sent to them (over any interface, remember we are
running 8.1 ...) we can probably safely use IGRP to pass around a
dynamic default route. It would be nice to use the PACCOM network as
our "default-network" but that would mean that if the satellite link
went down then ACT and NSW would not have a route back to the national
hub. We could of course pass routes for all Australian network numbers
to ACT and NSW from the national hub, but we have just put in place a
whole bunch of filters to avoid this (and the issue of metrics when
redistributing RIP to IGRP appears again, but this is less of a problem
in this case since the other states are all still in a star topology).
ARE THERE ANY OTHER WAYS OF PASSING DEFAULT GATEWAY INFORMATION AROUND ?
Would this work ?
router igrp 1221
redistribute static
default-network 132.160.0.0
!
ip route 132.160.0.0 139.130.x.y
ie. advertise 132.160 via IGRP, marked as an "exterior route" but don't
rely on it being passed dynamically to us from the Proteon.
Finally, on a related issue:
DOES AN IGRP "EXTERIOR ROUTE" OVERRIDE A RIP DEFAULT NETWORK (0.0.0.0) ?
(I assume so because IGRP routes have a smaller administrative distance
than RIP derived ones by default).
Comments gratefully appreciated,
Peter Elford, e-mail: P.Elford@aarnet.edu.au
Network Co-ordinator, phone: +61 6 249 3542
Australian Academic Research Network, fax: +61 6 247 3425
c/o, Computer Services Centre, pager: +61 6 245 3035
Australian National University post: PO Box 4
Canberra, AUSTRALIA Canberra 2601P.Elford@aarnet.edu.au (Peter Elford) (05/13/91)
On Thu Apr 18 10:54:17 1991 I wrote: >... > >We are about to install two E1 circuits as shown below: > > ACT NATIONAL > +---------------+ 2Mb +---------------+ > NSW | 139.130.188.2 +---------+ 139.130.188.1 | >+---------------+ 2Mb | | | | >| 139.130.192.2 +---------+ 139.130.192.1 | | | >| | 48K +---------------+ | | >| 139.130.72.2 +-----------------------------------+ 139.130.72.1 | >+---------------+ +---------------+ > > >The 48K between ACT and the national hub will be retired when the 2Mb lines >go in, but the 48K between NSW and the national hub will be retained for >backup. > >HOW DO WE SET UP OUR ROUTING TO DO DYNAMIC FAILOVER TO THE 48K IF ONE >OF THE 2Mb LINKS DIES ? This is a summary of the responses I got and our eventual solution; many thanks to those that answered. >My current thinking goes as follows: > >A) Stick with RIP. We can only do this if there is some method of >artificially increasing the hop count of routes sent from NSW to >the national hub over the 48K line. The "offset-list" parameter only >allows you to fiddle routing metrics on a per network, rather than >per interface, basis. > >IS THIS TRUE ? CAN WHAT I WANT BE DONE ? It cannot be done using RIP metrics. It is possible to do it using cisco's administrative weight "meta-metric": The NSW hub assigns an administrative weight of 254 to routes coming from 139.130.72.1 (the national hub). The national hub assigns an administrative weight of 254 to routes coming from 139.130.72.2 (the NSW hub). Since 254 is the largest possible value (about from 255 which says never believe), neither NSW nor the NATIONAL hub will believe the routes the they receive from each other, unless they are the *only* routes they receive (which is what happens if one of the 2Mb lines goes down). It takes about 3-5 minutes for the failover to the 48K to work because of the various timers within RIP, but it recovers almost instantly. >B) Fire up IGRP to be able to dynamically detect the failure of one of >the 2Mb lines. NSW and ACT routes will have to distributed between all >three routers but since most of them have are learnt via RIP, to >redistribute them using IGRP we will have to use a "default-metric" >for them all. > >DOES THIS MEAN THAT ALL RIP ROUTES REDISTRIBUTED BY IGRP WILL HAVE THE >SAME METRIC REGARDLESS OF WHICH INTERFACE THEY GO OUT OVER ? Yes. Quite a few respondents said we should just IGRP, but for such a simple topology and a simple requirement this seemed (and was) unnecessary. >ARE THERE ANY OTHER WAYS OF PASSING DEFAULT GATEWAY INFORMATION AROUND ? Lots! We settled on a default network solution like this at the NATIONAL router router rip redistribute static default-network 132.160.0.0 ! ip route 132.160.0.0 139.130.x.y The ACT and NSW routers then just have default-network 132.160 and failover works the same as for any other network. >DOES AN IGRP "EXTERIOR ROUTE" OVERRIDE A RIP DEFAULT NETWORK (0.0.0.0) ? Yes. Peter Elford, e-mail: P.Elford@aarnet.edu.au Network Co-ordinator, phone: +61 6 249 3542 Australian Academic Research Network, fax: +61 6 247 3425 c/o, Computer Services Centre, pager: +61 6 245 3035 Australian National University post: PO Box 4 Canberra, AUSTRALIA Canberra 2601