Reading about IS-IS now - lots of similarities with OSPF.
One of the topics the curriculum discussion, is IS-IS will forward a packet destined for inter-area travel to the nearest L1-2 router. From there L1-2/L2 routers will route to packet to the destination area and to the closest L1-2 router that directly connects to the area. Upon the return path, packets may flow a different direction because the closest exit L1-2 router may not be the same as the entry L1-2 router that delivered the packets to the destination area. I guess the chapter assumes there are multiple exit points (L1-2 ISs) in the area much like ABRs in OSPF. I have no problems with this concept as every router makes its own decisions based off its routing table.
Does OSPF and EIGRP work the same way?
The curriculum never talk about information taking a different path upon returning in EIGRP and OSPF. My impression after thoroughly reading the workings of OSPF, was that link state routing protocols held so much information about link states, cost, and shortest path, that all routers would make identical decisions since their LSDBs area all identical (Again, I understand that even with identical LSDBs, every router will make their own decision in their position on the network). Therefore the shortest path in one direction *should* be the shortest path in the returning direction. If it is possible for OSPF and EIGRP to perform like IS-IS in the routing logic, why wait until IS-IS chapter to describe a characteristic of most routing protocols?
Thanks - Mike
In this picture, I have setup a multiarea OSPF domain for the purposes of demonstrating if OSPF would perform asymmetric routing like IS-IS by forwarding packets destined to other areas to the nearest L1-2 IS (ABR). As you can see area 2 and 1 are connected to area 0. All routers are converged with no summarization or specifics.
The experiment was to ping from the physical serial interface of Area2 router in Area 1 to Area0 router in Area 2. (Sorry for the confusing labels.) The thought was that the ICMP pack would travel from Area2 to its nearest ABR (ABR2)->ABR0->Area1->Area0. Then, the return path would be from Area0->ABR1->Backbone0->Backbone1->ABR2->Area2.
When running the simulation and watching the flow of the ICMP packet, OSPF was dissimilar from IS-IS in that Area0 router tried to forward the packets destined for another area to its nearest ABR. Instead it flowed reciprocally back to Area1-> ABR0->ABR2->Area2 router.
Therefore, OSPF does not suffer from the same asymmetric routing problems that IS-IS might because it doesn't have the "forward to the nearest exit point" rule.
Can anyone confirm this finding?
Thanks - Mike
I think you are over-thinking this. Just from the reading...
First, keep in mind that every router makes its own local decisions based on (whatever) criteria happen to be in place. There's never a guarantee of symmetrical routing decisions.
ISIS L1 routers receive routes with an 'attached' bit set from the L1/L2 routers that have connections to another L2 device. The L1 routers pick the best one and create their own 0.0.0.0/0 route pointing towards it.
At the same time though, L1 routes ARE sent up through the multiple L1/L2 routers in detail. So different (external L2 routers) may make independent decisions about which "ABR" (loosely stealing this) to send return traffic on. Nothing to do with the other direction's choices.
OSPF isn't any different for this. ABRs will send information in and it will have costs associated with them (metrics) at which point area routers choose the best ABR to use to send out. Routers in other areas choose the best path as well, but from a different perspective, so it may be a completely different path.
One of the key enhancements to OSPFv2 was "transit capability" which meant that not every packet destined for another area actually had to transit area 0. LSAs do, but if an ABR knows of a "shortcut" it could take it.
IS-IS in a pure environment doesn't exactly have the same concept. If you ran all L2 connections, you may see some similar behavior, but it would depend on the details.