On (2007-06-21 16:34 -0500), Justin Shore wrote:
> Many thanks for the input, Robert. That's close to what we're doing as > well. We also have the prc-interval set to 10, increased > max-area-addresses and set NSF to cisco (plus some evil redistribution > that is temporary and will be gone soon). "ip fast-convergence" isn't > an option on our 7600s. What platform are you working on? I'm going to It is, it was just renamed to 'fast-flood x' in recent IOS'. What it does is send x amount of LSPs before counting SPF itself first. Counting SPF can take anywhere from 5ms to hundreds of ms's, and if in each hop you first calculate SPF then flood, you're going to have quite bad convergency time. Other things to tune in IS-IS for convergency is iSPF, route priority tag (put it to critical subset of your IS-IS speakers) and spf/prc/lsp-gen intervals (single most important thing would be not to wait IOS standard 5500ms for 2nd LSP before starting SPF), Then slightly related is to make sure your interfaces go down fast, here 'carrier-delay msec 0' and 'dampening' in interface config helps, however I've been told that some transceivers do not 0ms delay. If you don't have hardware livelyness detection you should use BFD which doesn't work that great in 7600. For initial IS-IS convergency (which shouldn't actually be very important) you can increase the speed in which LSP's are flooded by setting 'isis lsp-interval' in interface config. On ethernet you should set all links when possible to 'point-to-point' to avoid creating pseudonode (SPF complexity depends on amount of nodes and links) and to have reliable LSP flooding instead of periodic flooding. If you have your links in iBGP or you simply don't need them there, consider setting 'advertise-passive-only' also. Thanks, -- ++ytti _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
