> Although there is some discussion on it later there is no discussion > of the number of addresses to be learned in the single and multi-area cases and > the impact this has on the LSDB. The number of addresses to be learned will > impact the ingress RBridge FIB and the FIB update time so this is just as > significant in understanding the benefit of multi-level as understanding the > link-state convergence time is. The number of MAC (actually MAC/VLAN or MAC/FGL) addresses to be learned by an edge TRILL switch is not affected by whether the TRILL campus uses single or multilevel IS-IS. This number of MAC address learned is the seventh scaling problem listed in Section 1.1 which states that multilevel TRILL IS-IS helps only with the other six problems. This point, that multilevel doesn't help here, could be emphasized more but I don't think I see much point in this document in going into details such as the scaling considerations of data plane MAC address learning (which is the default in TRILL) or control plan MAC address learning (which does not use the core LSDB but rather Data Label scoped ESADI link stat databases). SB> I think the important point is either to demonstrate that the SB> that the control plane problem you address is the dominant factor in convergence SB> and that the MAC component (including FIB update) is not significant, SB> or to point to a solution that makes the data plane element comparable SB> to the improved control plane. Otherwise there is surely no point in SB> making these significant changes to the control plane. - Stewart |