Hello,
Please allow me to join and add my two cents...
1) Why would anyone use `redistribute connected`?
You don't see this command used often, and you are right that the network command is usually better suited to inject directly connected networks into EIGRP. As Paul correctly pointed out, one of the crucial differences between a network command and a redistribute connected command is that networks added with a network command are treated as EIGRP internal routes and EIGRP will try to establish adjacencies over the respective interfaces by sending and receiving EIGRP messages, while networks injected using the redistribute connected command are treated as EIGRP external routes and EIGRP will never send or receive its messages over such interfaces.
The redistribute connected would be advantageous in the following scenarios (the list is not exhaustive at all):
- Sometimes, you have a scenario in which you do not precisely know what IP address a network interface is going to have. For example, you can have some of your router's interface addresses assigned using DHCP or PPP IPCP. In those cases, you can hardly write a network statement to catch such IP networks. If you only want to have these networks advertised in EIGRP but not to create adjacencies over such interfaces, you're perfectly fine with redistribute connected. This might be specifically useful on routers that aggregate and terminate PPPoE sessions from DSL customers where customers are assigned potentially different IP addresses (some have a pool of temporary addresses, some others may have a range of permanent addresses).
- Sometimes, you are using a configuration template to be applied to many different routers, and you do not want to customize that template for each and every router. An easy way for this template to have all directly connected networks injected into the routing protocol is to use redistribute connected - this would be a common setup of Provider Edge (PE) routers in an MPLS Layer3 service to inject the PE-CE (provider-customer) links into BGP, without knowing beforehand what addressing is in use.
2) Does `eigrp network` always take precedence over `redistribute connected`?
To my knowledge, it does, and there is a good reason for it: In EIGRP path selection logic, internal routes are preferred to external routes (why would you go to a destination network over some foreign network topology if you have an internal path to it already?). So if you have covered a single directly connected network both with a network command and a redistribute connected command, then EIGRP will prefer the "internal point of view".
3) If EIGRP default-metric is not set, what metric is used?
For redistribute connected, EIGRP will have a look at the component metrics (bandwidth, delay, optionally reliability and load) of the interfaces where these networks are connected, and will compute the resulting metric out of these values. This is, by the way, precisely the same thing EIGRP would do if you had the networks added by a network command.
The default-metric is required specifically for redistributing routes from other routing protocols. Neither redistribute static nor redistribute connected in EIGRP require an explicit redistribution metric to be specified.
One last comment about passive-interface command: In fact, a passive interface neither sends nor receives routing protocol messages. This is true for EIGRP, OSPF, and IS-IS. RIP is the only exception - with RIP, passive interfaces do not send updates but they still listen for them.
Best regards,
Peter
