Help! Confused with split horizon

gbadman

I was working my way through chapter 2 of the Route Foundation Learning Guide and I basically hit a wall with the DUAL example. Here is the page on safarionline (page 75): http://my.safaribooksonline.com/9780132550376/ch02lev1sec2#X2ludGVybmFsX0ZsYXNoUmVhZGVyP3htbGlkPTk3ODAxMzI1NTAzNzYvNzU=
The page should be fully visible without subscription.

Fig.2-9 shows the topology tables on routers C, D and E. I have a number of issues with this example, which may well be due to my incomplete understanding of DUAL and split horizon.

1) The reported distance of A for the subnet seems to be ignored in the calculated FDs and ADs. Unless of course it is included in the metric (1) for B.
2) Given that D and E are present in C's table, does it then not follow from split horizon that only B should be present in D's table? In the example, C is also present in D's table.
3) Also from split horizon, I expected only D to be present in E's table. In the example, C is also present.

Running the topology in Packet Tracer (only K3 set and delay set to 1 and 2 accordingly) seemed to confirm my expectations, but showed other behaviour that cast doubt on that. C failed to show E in its topology table even with the all-links option. And E doesn't have C in it's own table either, so the culprit doesn't appear to be split horizon.

Raptien

1) It has likely been omitted for simplicity, it's not important for the example.

2) No because the route that C advertises to D is via B, so it wasn't learnt from D and can therefore be advertised to it

3) This is the way it should be, for much the same reasons as 2

gbadman

Ah yes. Duh. C learned the destination from D, not the route.
Thanks Raptien.

Raptien

I'm not too sure what you mean by 'C learned the destination from D, not the route'. In terms of routing entries, route and destination are kind of one in the same.

gbadman

Well there is a distinction. You can have multiple routes to the same destination. The subnet is the destination, the path (A-B-D-C etc) is the route. Split horizon prevents the feedback of routes, not of destinations.
So a router can feedback a destination it learned from a neighbour back to that neighbour, so long as the route it presents is different from the one it learned from the neighbour.
That's what I didn't appreciate until your explanation.

I suppose the statement 'C learned the destinationfrom D, route' is the wrong way round. It's more accurately 'D learned the destination from C, not the route'. I'm guessing that's what threw you.
D did indeed learn a route from C, but a different one from the one it advertised to it.

Raptien

It's just a matter of terminology I think, but your last sentence is spot on.