Name | IGP or EGP | Type | VLSM | Updates | Algorithm | Metric | Convergence | Scalability | Configuration | Port/Protocol |
---|---|---|---|---|---|---|---|---|---|---|
RIP | interior (intra-AS) | distance vector | classful | broadcast full routing table periodically | Bellman-Ford | hop count | slow | poor | easy | UDP 520 |
RIP-2 | interior (intra-AS) | distance vector | classless | multicast full routing table periodically | Bellman-Ford | hop count | slow | poor | easy | UDP 520 |
IGRP (obs.) | interior (intra-AS) | distance vector | classful | ?broadcast/multicast full routing table periodically | constrained bandwidth and cumulative delay (optionally others) | slow/med. | good | medium | IP 9 | |
OSPF | interior (intra-AS) | link state | classless | LSA. partial updates to topology to known neighbors | Dijkstra shortest paths | sum of links' cost (cumulative bandwidth) | fast | good | hard | IP 89 |
EIGRP | interior (intra-AS) | balanced hybrid | classless | partial updates of topology to neighbors | DUAL | constrained bandwidth and cumulative delay (optionally others) * 256 | (very) fast | good | medium | IP 88 |
BGP | exterior (inter-AS) | classless | list of ASNs to path to peers | none? policies | TCP 179 |
Notes: IGRP and EIGRP are Cisco proprietary.
EIGRP also classified as distance vector. Has replaced use of IGRP.
what to send | when to send | to whom | can form Loops? | |
---|---|---|---|---|
Distance vector | routing table | periodically | neighbor routers | yes |
Link state | my links status | when topology (i.e. my link status) changes | flood to all routers-->fast convergence | no |
Distance vector:
router/link failures can create instability/loop:
maximum distance is 16.
split horizon: don't send to router a route learned from that router
(i.e. don't tell someone something you learned from him).
poison reverse: override split horizon to send a "net unreachable"
to router you learned the route from (i.e. if you hear bad news,
pass it to the person you had been hearing the good news from).
If internet has cycle, still can have count to infinity upon
failure.
holddown interval: ignore updates about a route learned to be
unreachable to give time for the other routers to learn about it
and adjust (this update might be stale news) (i.e. ignore what
others say because they might not have heard the bad news yet).
negative reachability updates to obviate long timeouts of holddown
timers.
triggered updates so don't wait until next periodic update to send news.
Not so simple after all...
redundant links or parallel paths | require VLSM or discontiguous major networks | simplest |
---|---|---|
no | no | RIP V1 |
no | yes | RIP V2 |
yes | no | IGRP |
yes | yes | OSPF or EIGRP |
OSPF
within an AS (autonomous system), all Areas connect to Area 0 (backbone) at an ABR (area border router).
all routers in an Area have the same topology table (updates localized to Area).
only ABRs and ASBR do route summarization.