 |
 |
On Aug 21, 2012, at 14:07 , Doug McIntyre <[log in to unmask]> wrote: > On Tue, Aug 21, 2012 at 03:54:19PM -0500, Jay Hanke wrote: >> On Tue, Aug 21, 2012 at 3:46 PM, Mike Bushard, Jr >> <[log in to unmask]> wrote: >>> I typically would agree Owen, but in this case it may make sense. Everyone >>> will be down already, and we won't have multiple outages to the entire >>> exchange for IP changes. There will be a considerable amount of work if >>> everyone changes addressing on different schedules. I think just saying >>> that on day x we are on the long term strategy will save time in this >>> case. Our topology is pretty basic today so I think the rick would be >>> minimal at this time. >> >> The time frame is pretty tight for everyone to coordinate bilateral >> changes. Esp for the bigger folks. >> >> What about enabling the new IP addresses on the existing topology and >> moving one of the route servers to the new block. Then setting a flag >> date to remove the existing IP addresses 2 months out. > > > I'm sorry I've been so under the weather, that I've been behind and > haven't started coordinating the renumbering with my co-conspirators. > > Jay, one problem with a method like that, is that there isn't a > central router. Everybody is talking all in the same flat space. If > somebody changes to talk on their 206.108.255.0/24 IP address, then > those that haven't changed yet won't know how to reach their routes. > Thus breaking the exchange apart into those that have adjusted, and > those that haven't adjusted. What more, the unconfigured people will > see the new routes, but can't reach them. > There's no reason people can't add the new addresses and migrate peering sessions individually, one peer at a time, then remove the old addresses once that's complete. I think two months would be a bit narrow on the time frame, however. > > What follows is the plan I had in my head (but not communicated yet) > until I saw the oppertunity to just do it all at once. > > We communicate everybody's new 206.108.255.0/24 IP & 2001:504:27::/48 > address, which almost certainly will just be reusing the same last > octet as the current one. > Might I suggest on the IPv6 block 2001:504:27:0::/64 be used for the exchange (we can use other parts of the /48 later if needed) and that we number as follows: 2001:504:27:0:AS:NO:XXXX:YYYY Where AS:NO is the 32 bit as number and XXXX:YYYY is a number chosen by the applicable AS to represent the specific router/port combination. (Yes, 32 bits for router/port combinations is overkill, but I can't think of a better use for the bits). > Everybody does a secondary of these IP addresses on their MICE facing > interface. This has to be completed by such-and-such date. We'd do > spot checks with ping to make sure people have completed this. > I would actually suggest making the existing address "secondary" and the new addresses "primary" to reduce future downtime. > Meanwhile, we build duplicate BGP sessions in the route reflectors for > when people are ready to cut. > +1 > After the such-and-such date, then people are free to cut over their > BGP sessions to originate from their new IP, and connect to the new > BGP session on the route reflectors. > I think the RS peering sessions can be cut this way, but suggest you look over my earlier proposal for managing the peer to peer sessions. > Finally, there is a drop-dead date when all use of the old IP ranges are done. > > This way, when somebody cuts onto the new BGP session, the > unconfigured old people still have a way to reach the new routes > (via. the secondary IP addresses configured on their interface). But > their BGP sessions don't change until they want them to. > An excellent solution for the RS peering sessions. > This way is slow, and requires 100% compliance in getting the > interface secondary addresses, and then in the final drop-dead date. > Actually, it doesn't require 100% compliance. People can actually add the interface secondaries at their convenience and everything continues to work with their old address until they choose to cut over. > OOTH, if everybody goes down for the switch cut, then people can bring > up their session when they want to after the maintenance window, but > that also means everybody has to be available to cut their configs > over that night/weekend as well if they want to continue to talk on > the exchange. If everyone were only peered with the RS, I'd say this might be a feasible approach. However, since they aren't, I'd suggest that we point out the switch cut as a "recommended time" to apply the new addressing to their interfaces and leave it at that. Owen ######################################################################## To unsubscribe from the MICE-DISCUSS list, click the following link: http://lists.iphouse.net/cgi-bin/wa?SUBED1=MICE-DISCUSS&A=1