On Aug 22, 2012, at 14:51 , Doug McIntyre <merlyn@IPHOUSE.NET> wrote:
On Wed, Aug 22, 2012 at 02:47:21PM -0500, Justin Krejci wrote:
While I agree with the general sentiment that the IP change and switch change will likely be smooth, the IP change can be done more transparently and has literally nothing to do with the L2 gear. To that end, I don't see any reason to not configure/assign the new IP block now and let folks start using it right away, even before the switch switch.
But, we don't have a central router?
If your border to the exchange right now got a prefix with the next hop of 206.108.255.3, how'd you get there? You'd have no route to host for 206.108.255.3.
That is why I brought up the secondary IPs. Once everybody puts in a secondary IP in 206.108.255.0/24, they'll know how to try to get to 206.108.255.3 for my prefix announcement. But I can't really do the prefix announcement with the next-hop in 206.108.255.3 until everybody knows how to get to 206.108.255.0/24?
Your next hop for peers that are still peering on the old address should be your old address. For peers on the new address, should be your new address. The problem you describe could only apply, therefore, to the routeservers which are providing next hop data other than their own local interface. In that case, there are multiple possibilities: 1. Run two route-server instances on each address set. 2. Run one route-server instance on each address set. The route-server instances on the old address set shouldn't be seeing new address next-hops. The route-server instances on the new address set shouldn't be seeing old-address next-hops. Yes, it means everyone that is using new addresses has to maintain their old sessions with the route-servers until everyone has a new session. As Mr. Tindle said earlier... I've been through a few of these. They're not rocket science, but trying to herd all the cats into doing it as one massive change all at the same time is an almost certain path to large-scale disruption and lots of time debugging stuff that didn't have to be debugged all at once. A little careful planning and adding sessions one at a time and then judiciously removing sessions that are no longer needed along the way can accomplish this in a reasonable timeframe with a minimal amount of fuss and no mass-scale risks. Owen ######################################################################## To unsubscribe from the MICE-DISCUSS list, click the following link: http://lists.iphouse.net/cgi-bin/wa?SUBED1=MICE-DISCUSS&A=1