For a network to converge, all routers in the network must collect and agree on all the topology information from each other. This information must be consistent, reflecting the current state of the network and free of routing loops or any other kinds of corruption.
What is meant by fast convergence?
when we use the word fast we are speaking about a relative thing specfied in time measuring units. The word fast here is describing the convergence time, which is the time required by all the routers in the network to flood and use the topology information. By fast convergence we usually mean sub-second convergence time.
Why do we need that level of convergence? aren’t routing protocols default timers just fine?
In today’s networks; network interface speeds are growing rapidly and its common to see a 10 Gbit link between two devices specially in the core networks of service providers. Consider the failure of such a trunk for only two seconds and imagine how much information will be lost during the failure before the network is fully converged.
Default routing protocol timers work fine in networks that do not need such level of convergence speed and for lower speeds interfaces.
So, How do we achieve such level of convergence?
Actually there is a full portfolio of features and technologies on modern routers invented specially for the sake of improving convergence speed. As a network designer you need to get familiar with these technologies and their implications in order to correctly utilize them and keep your network stable. Generally these technologies address one of the following parts:
- Physical layer: Failure detection speed.
- Routing protocols: event reporting, other tables calculations speed.
- Forwarding: how fast the forwarding engine can react to topology changes and adjust its tables.
Some of these features are:
- Fast hello for OSPF and IS-IS.
- Bidirectional failure detection (BFD).
- NSF with SSO or graceful restart.
- Partial and incremental SPF calculations.
I hope I have made some of the concepts clear and We are going to discuss of these features implementations and best practices in later posts.