The Complete IS-IS Routing Protocol- P9

The Complete IS-IS Routing Protocol- P9:IS-IS has always been my favourite Interior Gateway Protocol. Its elegant simplicity, itswell-structured data formats, its flexibility and easy extensibility are all appealing – IS-ISepitomizes link-state routing. Whether for this reason or others, IS-IS is the IGP of choicein some of the world’s largest networks. Thus, if one is at all interested in routing, it is wellworth the time and effort to learn IS-IS.
Nội dung trích xuất từ tài liệu:
The Complete IS-IS Routing Protocol- P9 IS-IS Application Level Fragmentation 2459.5 SummaryContrary to IP routing protocols, IS-IS cannot rely on a network layer to do fragmentationfor it. IS-IS runs directly on the link-layer, which has no possibility of fragmenting frames.IS-IS therefore needs to apply a few techniques to get around the too-small MTU problemif it has to transmit a message that is larger than the MTU. The two techniques IS-IS usesfall under the category of fragmentation avoidance and application level fragmentation.IS-IS assumes a minimum MTU that every link has to support. This limit today is theoreti-cally 1492 and in practice 1497 bytes. Additionally in several packet types in IS-IS thereis support for multi-packet messages like CSNPs and LSPs. There are also details of howthe application IS-IS should fragment in order to avoid network-wide churn. “Cluey”implementations think in terms of fragments and try only to rebuild fragments that havebeen affected by an adjacency change. In current network architectures, the distributedLSP storage space is typically only utilized at 10 per cent, and even if the space couldbecome exhausted, the IS-IS working group has come up with a solution that is similar tomodelling adjacencies on a large LAN. There is also the possibility of raising the 1492byte limit of LSP buffer size as all modern interface cards, especially in core environ-ments, support MTUs up to 4474 bytes. Although not needed today, these developmentsare a good proof that the IS-IS infrastructure will exist for a long time to come in networks.8Synchronizing DatabasesLink-state protocols rely fundamentally on the fact that each router in a given area hasthe same view of the topology. Sharing the same view is the foundation for computingconverged routes. Convergence means that each router computes routes in a way that movespackets one hop closer to the destination. If routes are not convergent, packets can takeextra hops in the network to reach the final destination. The worst case for misalignedroutes is a forwarding loop where packets are bounced back and forth between a pair ofrouters. In a changing network it is therefore paramount to get to a synchronized viewof the topology as quickly as possible. New routers connected to the network need toquery their neighbour’s link-state databases to get on the same page as quickly as pos-sible. This chapter discusses the IS-IS mechanisms to initially synchronize link-statedatabases between routers and periodically update the databases. This chapter starts with a detailed discussion of the consequences of link-state data-base misalignments. It continues with a detailed explanation of database synchronizationon point-to-point and broadcast links. Finally, refinements to ISO 10589 for increasingthe robustness of the link-state database synchronization process are presented.8.1 Why Synchronize Link-state Databases?Unlike distance vector or path vector protocols, such as RIP or BGP, which compute theirroutes by travelling through the network, link-state protocols make a local computationbased on shared and identical topological information. To contrast this difference better,let’s consider the following example: Figure 8.1 shows how RIP calculates the distance to anetwork. Router A is locally connected to the sub-net 192.168.1/24. The router is distribut-ing its reachability information to the prefix 192.168.1/24 by sending a RIP update to all ofits neighbours (Router B and Router C) with a metric of 1. Routers B and C distribute theinformation further to other neighbouring routers by incrementing the metric field by one.Routers D and E, which receive the update, install the prefix 192.168.1/24 with a metric of2 in their routing and forwarding tables. Observe that the actual calculation of the metrichappens in a distributed fashion: Routers D and E do not know where the prefix origin-ated. All they know is that it is two routers away and that Routers B and C have receivedthe prefix. So Routers D and E do not have any topological visibility with regard to the pre-fix. Each router modifies the original advertisement by incrementing the metric field. Link-state protocols, like IS-IS, work differently. Each IS-IS router in a given routingdomain has to generate its own view of the local network to its neighbours. Next, all these 205206 8. Synchronizing Databases 192.168.1/24 192.168.1/24 via ETH 0/0 Metric 0 Router A 192.168.1/24 192.168.1/24 via Router A via Router A Metric 1 Metric 1 Router B Router C 192.168.1/24 192.168.1/24 via Router B via Router C Metric 2 Metric 2 Router D Router EFIGURE 8.1. RIP sends out information about local networks plus remote networks by incrementingthe metric fieldindividual, local views are passed on to the other routers. No router is allowed to change theoriginal information. Ultimately, each rout ...