Key points that I only understood fully when studying for the CCNP routing exam this time around:

Redistribution:

Routes are redistributed into the destination protocols topology table/database, no the Routing Information Base (RIB).

Routes must have made it into the RIB from the source protocol to be considered for redistribution into another protocol.

Connected networks for which the source protocol is enabled are redistributed also (IPv6 routing protocols do not do this unless “include-connected” is used).

Redistribution of non-classful routes into OSPF requires the “subnets” keyword on the redistribution statement (OSPF v3 for IPv6 does not require this).

Configured PBR over-rides the default routing paradigm (destination-based forwarding). If the “set” action specifies the keyword “default” before the action, PBR is used if no destination-based decision can be made from the RIB.

If no metric is specified for routes redistributed to the destination routing protocol RIP and EIGRP will give the routes an infinte metric, making them ineligible for use.  OSPF defaults to a cost of 1 routes redistributed from BGP, the original cost for routes from another OSPF process, and 20 for routes from any other IGP.  BGP uses the IGP metric for the route as it’s metric.

Default administrative distances can be tweaked to make certain types of routes in a protocol more or less favourable compared to route types in other protocols.  The specified distance applies to the current router only:

router rip
  distance  (can optionally include a   for selective tweaks)

router eigrp x
  distance eigrp  

router ospf x
  distance ospf external  / intra-area  / inter-area

Distribute-lists for route redistribution can only be used “out” and filter the eligible routes passed into the destination protocol.  Since OSPF requires the LSDB to be the saem for all routers in an area, it allows the use of distribute-lists “in” to filter between the OSPF database and the RIB.

Filtering with route-maps:

A route-map permit line, with a successful match in the match statement = ALLOW the prefixes (with optional set).

A route-map deny line, with a successful match in the match statement = FILTER the prefixes.

Whatever the match mechanic in the route-map is (access-list, prefix-list, as-path access-list), it is just used for matching, not for filtering. A deny in the match method just means this isn’t a match, and allows the prefix through to the next step.

Route-maps have a non-printing implicit deny statement at the end.

Route-map match logic:

match 10 20

Means match 10 OR 20 (as in either or both of 10 and 20 count as a successful match).

match 10
match 20

Means match 10 AND 20 before considering this a successful match.

Using an extended access-list to match prefixes (you just wouldn’t in real life!):

To match an exact prefix (e.g. 10/8 only) the syntax is either of:

access-list 101 permit host 10.0.0.0 host 255.0.0.0
access-list 101 permit 10.0.0.0 0.0.0.0 255.0.0.0 0.0.0.0

To match a range of prefixes  (e.g. any route starting with 10. in the first octet and a 24 bit mask):

access-list 101 permit 10.0.0.0 0.255.255.255 255.255.255.0 0.0.0.0

Summarization:

The metric assigned to a summary address is always the lowest metric of any of the prefixes that fall under the summary address, with the exception of OSPF v3 which uses the highest (OSPF v2 can be made to do so also).

Even with auto-summary on, a routing protocol capable of sending the mask with the prefix (i.e. all but RIP v1), will send the real prefix/mask in updates, converting to classful summaries only when crossing a classful boundary (e.g. if an EIGRP network uses 10.x.x/24 throughout and auto-summary is on, all routers will have all 10.x.x/24 prefixes in their topology tables.

RIP:

• V1 is classful (it cannot send the prefix mask with the prefix in updates) and broadcasts updates.
• V2 is classless, but defaults to auto-summary, and multicasts updates.
• NG is classless (IPv6 version).

V2 passive-interface prevents the sending of updates but allows the receipt of updates.

A static neighbor statement turns off multicast and causes updates to be unicast to the defined neighbor only.

EIGRP:

Feasibility condition: A route is considered a feasible successor if it’s AD is less than the current successor FD.

FD is the metric to get from this router to the destination. AD is the metric that a neighboring router reports to get from itself to the destination.

“2 successors” means that there are two equal cost paths to the destination.

EIGRP can do unequal cost load balancing via the variance and max-paths (default=4) command, but ONLY if the alternate paths are already feasible successors. To work out which paths qualify:

1. Work out which alternate paths are feasible successors (AD < FD).
2. Work out the metrics of those paths.
3. Eliminate any of those paths whose metric exceeds variance multiplied by the FD.

EIGRP stub are responsible for announcing they are stubs (via hello) to neighbors. Connected and Summary are the default prefixes that an EIGRP stub will advertise, other options are; Static, Redistributed, and Receive-Only.

EIGRP’s “ip default-network” command is classful and requires a full match in the RIB.

EIGRP neighborships don’t require matching values for anything other than which “k values” will be used (the actual value of the k value in question does not need to match e.g. neighbors can have different delays for the same link).

A router goes active on a route if it has no feasible successor when the successor route is lost.

Query floods stop at a router with the route in passive state. Queries are not sent to stubs. Queries to a router with  a summary for the route but not the specific produce an immediate reply.

OSPF:

OSPF does not do unequal cost load balancing, only max-paths (default=4) equal cost load balancing.

All routers in an area must have the exact same LSDB contents (hence flood behaviour).

SPF only runs for changes to type 1 and 2 LSAs in the router’s area.

When configuring totally stubby areas, only the ABR requires the no-summary keyword (because it is the one that creates the default and blocks type 3 and 5 LSAs going in to the area). All routers within the area require only the area stub command.

In order to form an OSPF neighborship, everything bar the actual metric needs to match between neighbors.

BGP:

A BGP router only sends it’s best “>”  path in updates.

To advertise via “network” a full RIB match is required. To advertise via “aggregate-address” only a component in the RIB is required.

Soft-reconfiguration inbound increases memory overhead by storing a copy of the prefixes received from a neighbor, so that a subsequent clear with the word “soft” in can run the original prefixes through whatever new filtering logic has been applied. Route-refresh does not have a memory hit and just requests a new copy of the prefixes from the neighbor without upsetting the connection state. Without either, a clear will tear the connection down.