Do you agree with the design of OSPF?

CriscoSystems · ‎02-27-2009

A couple weeks ago I experimented with multi-area OSPF and was confused by the output of my "show ip ospf database" command. I didn't like the way some of the "Link ID's" were network ID's and some were router ID's.

Everyone here was very helpful and patient with me (thanks) and I think I have made my peace with the simple (but for my purposes, still illogical) fact that only multiaccess networks will have their network ID's appear as Link ID's within their own area. Because intra-area network ID's come from Type 2 LSA's. And only DR's, which of course are found only on multiaccess networks, get to create Type 2 LSA's.

OK fine; so that's how it is.

But is anyone else out there at least FRUSTRATED, like I am, by the fact that that's how it is? Point-to-point networks get to have their network ID's appear in other areas as summary (Type 3) LSA's!! Why I ask you WHY should this convenient type of listing be denied to same-area routers?!?!?

Why I ask you WHY shouldn't a point-to-point router be allowed to create a Type 2 LSA for its perfectly respectable point-to-point network?

ssssiiighhh... like I said, I know I have to make peace with this; but I'll get there sooner if I know there's someone out there who agrees with me.

Or who can otherwise explain the reasoning behind the rules as they are.

(I never thought I'd say, "Life was simpler under BGP...")

Laurent Aubert · ‎02-27-2009

Hi,

LSA Type 3 is not related to the type of the link. They just include prefixes and netmask.

The advantage of not propagating LSA Type 1 & 2 outside the area is to keep the Tree calculation within the area for scalability reasons.

LSA type 1 & 2 are used by the routers within an area to build the map of the network which is of course the same on all the routers.

The link-ID of a pt-2-pt network is set to the router ID which is on the other end of the link (from the router originating the LSApoint of view). This information is used during next-hop resolution when building the routing table.

The link-ID of a stub network (no OSPF neighbor on this link) is set to the prefix itself because we need this information to build the routing table.

I can understand your point of view because with OSPFv2, information needed to build the tree and the prefixes itself are mixed in the same LSAs and we don't need to know the prefixes to build the map.

With OSPFv3 and IS-IS prefixes information are carried on dedicated LSA/LSP packets.

I think there are some work on-going so OSPFv3 will be address-family aware and be able to support IPv4 as well.

Looks like your holy grail ;-)

HTH

Laurent.

JamesLuther · ‎02-28-2009

Hi,

With OSPF then every router within an area needs to build exactly the same SPF tree. Bearing that in mind then it makes sense that there needs to be different LSAs to represent different network types otherwise each router would build a different map/view of the network.

However OSPF routers only do an SPF calculation for their local area. For intra-area then it works more like a distance vector protocol. Therefore intra-area routes may be easier to read in the OSPF database but there isn't enough info in there to build an SPF tree.

The fact that OSPF gleans so much information about the network topology is the beauty of the protocol. Just be glad it's not RIP.

Regards

Giuseppe Larosa · ‎03-01-2009

Hello Sueth,

you would like to have more user friendly commands, but this is not part of OSPF protocol design, but rather a question of implementation.

Some aspects are misleading and you are not the first and you will not be the last to find problems in consulting the OSPF database.

OSPF being a link state protocol is primarily concerned with node identifiers, OSPF router-ids.

Because every node has to draw a tree between itself and every other node in the area.

Router LSAs are labelled with the node OSPF router-id.

Network LSAs are not built to advertise an IP prefix but to represent a multi-access segment where one, two or several OSPF routers connect.

By examining the Network LSA it contains pointers to the Router LSAs of connected routers.

A router that is far from the segment using this info can draw a complete picture of the OSPF area including the multiaccess segment: it can know how many routers and what devices are connected to the segment.

It is a topological information.

sh ip ospf database advertising-router OSPF-router-id can give you more information about the contents of a single Router LSA.

So creating a network LSA for a point-to-point link is not performed because is not needed at all: every router can understand what are the two routers that connect on the link.

Information from other areas is instead prefix-based in summary LSAs.

Hope to help

Giuseppe

CriscoSystems · ‎03-02-2009

"Network LSAs are not built to advertise an IP prefix but to represent a multi-access segment where one, two or several OSPF routers connect.

By examining the Network LSA it contains pointers to the Router LSAs of connected routers."

See, here's the part I'm still stuck on.

Even if you have multiple routers connected to the multi-access segment, isn't there a Type 1 LSA created by each of those routers; and isn't the Type 1 LSA flooded throughout the whole area anyway? Why then the need to "alert" other routers in the area that there's a multiaccess segment out there?

Or, is the book I read incorrect; and type 1 LSA's travel only to their adjacent neighbors and not throughout the area?

JamesLuther · ‎03-02-2009

Hi,

Type 1 & type 2 LSA's are flooded throughout the whole area. Every router in the area needs to be able to calculate the SPF tree based on these LSAs.

So imagine trying to draw a diagram of your network only using point to point links. Can't be done and so OSPF needs to be aware of multi-access networks so it can calculate the tree.

Remember OSPF isn't like a distance vector protocols that just know it's x hops one way and y hops the other. It actually calculates a full map of your network (SPF tree).

Regards

Giuseppe Larosa · ‎03-02-2009

Hello Sueth,

your observation is good: can the router LSAs be enough ?

The answer is no.

Router LSAs are flooded in the whole area but they are thought to be missing some information:

for example who is the BDR on the segment.

in the router LSA a multiaccess segment is classified as a transit network (link type 2) and the link-id field contains the ip address of the DR node on the segment (that is not the router-id, they can be and are typically different).

in the case of a transit network (multi access) the ip address of DR interface is actually a pointer to the network LSA that has the same link-ID and contain a list of Router-IDs of connected OSPF routers (just like the OSPF hello in the segment itself).

An OSPF hello sent out a lan contains a list of the ospf RIDs of all routers in the segment just to make a comparison.

So you see that the Network LSA actually provides additional information.

Instead in the case of point-to-point links the link-id in the router LSA contains the OSPF router-id of the neighbor.

And here you see that a router examining the router LSA or router R1 can place a connector to node R2 if R1 and R2 are connected with a point-to-point link.

The real objective of OSPF is to allow each router to draw a complete and detailed map of all links and nodes.

Hope to help

Giuseppe

Laurent Aubert · ‎03-02-2009

The network LSA are here to represent a multi-access network in the OSPF topology. They allow the routers to see this network as a pseudo/virtual node with several routers connected to it.

DR concept reduces number of adj on the segment thus limiting the routing information exchange.

Laurent.