> Uh right, thats why it is called a FORWARDING information base. Only the > data that is used for forwarding is installed in the FIB, but the point is > that it is pre-populated with forwarding data for the entire table rather > than a "cache" which is only populated on demand. If there is a less > specific route which changes but it is covered by more specific routes > which do not change, nothing in the forwarding table changes. The same > holds true for routes which are not selected as best path. > > The problem with historic non-prepopulated lookup methods is that the > first lookup to a destination which has never been encountered before must > take the "slow path" through the CPU, resulting in non-deterministic > performance, and Very Bad Things (tm) when random destination traffic > (think worms scanning the Internet for new victims) hits your network. By > prepopulating the FIB, get consistent lookup performance to any > destination.
I am not an specialist ... but ... and about the performance for: - IPv4 Multicast traffic - IPv6 unicast traffic - IPv6 multicast traffic - MPLS L2 and L3 traffic (unicast and multicast) How the XMR deal with all these traffic at the same time ... with high performance ? The main goal of the XMR family is to compete with CISCO CRS-1 and JUNIPER T-640, right ? The CRS-1 development show something about how to use the SWITCH FABRIC in different ways to forward unicast, multicast, ipv4 and ipv6 traffic. How XMR deals with it ? Thanks a lot, Giuliano
