On Fri, Aug 05, 2005 at 02:04:21PM -0700, Peter Boothe wrote:

On Fri, 5 Aug 2005, Sabri Berisha wrote:

Hi Peter,

Everything can break. We want to know what will break less.

http://soy.dyndns.org/~peter/projects/research/anycast/nanog/

So, since Daniel Karrenberg showed that there was at least a 1% difference between anycast and unicast reliability, and we showed that the median AS switches root clusters every 3 hours, anycast is still a net win for the median AS, even with TCP queries, and is probably a net win for everyone except the people who have severe extant route flapping issues.

So what I understand from your slides is basically that you acknowledge the fact that it is theoretically possible for the setup to cause temporary loss of connectivity (very minor packetloss) which may result in DNS queries to timeout, but that the impact of such an event has no real-life damage. I thought that it would be virtually impossible to get this actually working outside of a lab-environment due to continuous route-changes on the net. I stand corrected. -- Sabri Berisha, Juniper Certified - JNCIA #747 | Cisco Certified - CCNA email: sabri@cluecentral.net | cell: +31 6 19890416 http://www.cluecentral.net/ | http://www.virt-ix.net/