home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.sys.cisco
- Path: sparky!uunet!charon.amdahl.com!pacbell.com!sgiblab!munnari.oz.au!metro!dmssyd.syd.dms.CSIRO.AU!megadata!andrew
- From: andrew@megadata.mega.oz.au (Andrew McRae)
- Subject: Re: re: advertising static routes
- Message-ID: <1992Nov6.052714.18648@megadata.mega.oz.au>
- Originator: andrew@noah
- Sender: news@megadata.mega.oz.au
- Organization: Megadata P/L, North Ryde, Sydney, Aust.
- References: <720752998.10981@news.Colorado.EDU>
- Date: Fri, 6 Nov 1992 05:27:14 GMT
- Lines: 141
-
- I would appreciate some net.feedback on this matter; bear with
- this posting, as it may seem complicated, but the situation is
- fundamentally straight forward.
-
- We have been dealing with some cisco router config problems that reflect
- some of these issues that people have raised here, and part of
- our struggle has been that the manual is too sketchy when it
- comes to some basic examples and explanation of how to set some
- things up. The basic problem we saw is that when you use
- ip unnumbered addressing on serial lines, you have to play some
- nasty games to get RIP routing working, and even then you come across
- some insurmountable obstacles (these may be bugs, I don't know).
- We are running 8.3(3), and we have the following network
- configuration (sorry about the size of diagram, hard to make it smaller).
-
-
- Ethernet 192.55.2.x
- ---+-----------------------------------+------------- net1
- |IP 192.55.2.20 |IP 192.55.2.20
- | |
- | Ethernet 192.55.1.x |
- | --+----------------------------|-------+----- net2
- | |IP 192.55.1.20 | |IP 192.55.1.21
- | | | |
- +------------+ +-------------+
- | mgs1 | | mgs2 |
- +------------+ +-------------+
- | |
- | Serial 9600 link | Serial 9600 link
- | |
- +------------+ +-------------+
- | | Serial 9600 link | |
- | cgs1 |----------------------| cgs2 |
- +------------+ +-------------+
- |192.55.10.1 |192.55.11.1
- | |
- | Ethernet 192.55.10.x | Ethernet 192.55.11.x
- ------+---+--------- -------+---+-------------
- | |
- | IP 192.55.10.2 | IP 192.55.11.2
- +----------------+ +----------------+
- | Annex T/S | | Annex T/S |
- +----------------+ +----------------+
- | | | | | | | | | |
- | 2400 Baud asynch | 2400 Baud asynch
- | Compressed SL/IP | Compressed SL/IP
- | IP 192.55.100.1 | IP 192.55.101.1
- +---------------+ +---------------+
- | Remote | | Remote |
- | Telemetry | | Telemetry |
- +---------------+ +---------------+
-
- Essentially we have four routers, with the two MGS's on dual Ethernets
- for redundancy; they each have a serial line (all HDLC encap) to
- separate CGS's. The CGS's are triangulated so that any one serial line
- can fail and you still have a path anywhere. This is only a portion of the
- net; eventually there will be 12 CGS's, so to simplify the routing and
- addressing, we decided to use ip unnumbered addressing on the serial
- lines, where the serial line used the same IP address as the Ethernet
- the router was connected to; the idea being that each router/terminal
- server cluster was a single net, advertising a static route to the
- boxes on the SL/IP side of the terminal server. *BIG MISTAKE*!
-
- The problem with this is that the router will not broadcast a route
- to its *own* net on the serial line e.g. When cgs1 broadcasts routes
- via RIP to both serial lines, it *will not* include the network 192.55.10.0
- in its tables, and so nobody else knows how to get to that network. Nothing
- I did convinced the router to include the network in its routing updates.
- The only way I saw to overcome this problem was to install a
- static route in the routers informing it what the net number of the router
- it was connected to e.g. for cgs1:
-
- ip route 192.55.1.0 serial 0
- ip route 192.55.11.0 serial 1
-
- And cgs2 had:
-
- ip route 192.55.1.0 serial 0
- ip route 192.55.10.0 serial 1
-
- And all this worked, *except* when you failed one of the serial links.
- Again, by example, if the link between mgs1 and cgs1 failed, then
- mgs1 would stop advertising the route to cgs1, and hosts would change
- their routing tables to go via mgs2 (all well and good). Mgs2 would pass the
- packets to cgs2, which would route them to cgs1 - OK, the packet got there
- fine, but when a packet is returned, cgs1 still has a static route to
- 192.55.1.0 via serial 0, EVEN THOUGH THAT INTERFACE IS DOWN. The static
- route is *not* overridden even when the router has failed the serial line
- (and yes, the router *does* show that interface as down).
-
- There is some talk in the manual about using administrative distance to
- override static routes, and so I set up some distances; lo and behold
- when you fail the serial link, the routers *do* override the static route
- correctly, but when the link is restored the static route is not restored,
- so the dynamic route (no longer the best one) is never removed.
-
- After some discussion with the local cisco support (and hearing a few
- times `yes, it *should* all work'), we put a different IP address (with
- different net number) on all the serial interfaces, and removed the
- static routes to the interfaces. *This* works, but it makes the
- routing tables much bigger, and everthing looks more complicated.
-
- SPECULATION *ON*
-
- We are seeing two problems here, one a routing issue (possibly
- not a bug), and a static route override bug.
-
- Since the IP interfaces are unnumbered, and use the Ethernet IP address,
- the router thinks that the serial interface has the same network number
- as the Ethernet, and so it *will not* broadcast a route to a network to the
- interface that is addressed as that network number, just as it will
- not broadcast a network route to the Ethernet that network is on e.g
- if a router ethernet interface is 192.55.10.1, then it obviously
- will not include a route to network 192.55.10.0 in its RIP advertisement
- broadcast on *that* network. The difference is that with a
- serial line it is making a wrong assumption, as the router on the
- other end of the link does not exist on the same network number,
- unlike an Ethernet. The other router, of course, does not know how
- to get to that network because the first router is not advertising it, unless
- you configure it with a static route to that network pointing to the
- serial interface.
-
- Is this a bug? Possibly it is just a bad assumption, but a safe and
- conservative one.
-
- The other problem seems more straight forward. When you have a static
- route to an interface, and that interface goes down, then the static
- route is never overridden, except if you play games with admin. distances,
- and then the route is never restored once the link comes up. The frustrating
- thing is that the router *knows* the interface is down, but it still
- maintains that route in the routing table, even though the interface
- is down *and* an alternative path exists. This seems like a bug; if
- not a bug, then certainly a difference in opinion about how
- static routes to interfaces should operate under interface failure.
-
- Am I expecting too much from the routers?
-
- Andrew McRae inet: andrew@megadata.mega.oz.au
- Megadata Pty Ltd, uucp: ..!uunet!megadata.mega.oz.au!andrew
- North Ryde 2113 Phone: +61 2 805 0899
- NSW AUSTRALIA Fax: +61 2 887 4847
-