I've been going through a couple of the presentations referenced on the page and from searching NANOG. Wicked stuff.<br><br>Keeping state was my big concern. One paper explains the whole configuration well:<br><br>Building Nameserver Clusters with Free Software
<br><a href="http://www.nanog.org/mtg-0505/abley.cluster.html">http://www.nanog.org/mtg-0505/abley.cluster.html</a><br><br>Each server in a single cluster speaks a routing protocol to the router to advertise its availability. This router then has multiple /32 routes to the service address which it can perform equal cost load balancing. The document points out that routers should be keeping the same path for each flow (source/dest IP pair) which is easily doable.
<br><br>But what I really couldn't get was how to ensure that one transaction always goes to the same cluster, given multiple clusters. If the AS Path goes and changes (ie failed link between providers), its possible that the user will get sent to a different cluster.
<br><br>Best Practices in IPv4 Anycast Routing<br><a href="http://www.sanog.org/resources/sanog5-woody-anycast-v10.pdf">http://www.sanog.org/resources/sanog5-woody-anycast-v10.pdf</a><br><br>has a slide talking about the statefulness. The implication of the slide (p39) is that "it shouldn't matter, your application should deal with it". The interesting part is the last point:
<br><br>"Limited operational data shows underlying instability to be on<br>the order of one flow per ten thousand per hour of duration."<br><br>ie, the Internet is stable enough.<br><br>A bit of Googling led me to
<a href="http://www.isoc.org/briefings/020/">http://www.isoc.org/briefings/020/</a> which has some commentary on the need for anycast dns and other gory details of the DNS system. Then there's <a href="http://www.cs.ucla.edu/~vpappas/p/ss_sigmet05.pdf">
http://www.cs.ucla.edu/~vpappas/p/ss_sigmet05.pdf</a> "On the use of Anycast in DNS" which takes a brief look.<br><br>Finally, I was curious as to how distributed these clusters were, ie if I could see different clusters from my different peers.
<br><br>I peer with Allstream and Sprint in Toronto, and GT and Sprint in Winnipeg. Using the F root server (<a href="http://192.5.5.241">192.5.5.241</a>) as an example, the three providers all peer with AS30130 which is a Toronto based network of ISC, and then AS3557 which is ISC in California who originates the
<a href="http://192.5.5.0/24">192.5.5.0/24</a> prefix. <br><br>Sean<br><br><div><span class="gmail_quote">On 6/6/06, <b class="gmail_sendername">Bill Reid</b> <<a href="mailto:billreid@shaw.ca">billreid@shaw.ca</a>> wrote:
</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">This is the page I was talking about at tonight's meeting. Interesting<br>commentary on the development of a new protocol.
<br><br><a href="http://www.av8.net/IETF-watch/DNSRootAnycast/History.html">http://www.av8.net/IETF-watch/DNSRootAnycast/History.html</a><br><br>As always it was an interesting meeting. Thanks to Sean and John for presenting
<br>and to all who came out and participated in the lively discussion.<br><br>-- Bill<br>_______________________________________________<br>Asterisk mailing list<br><a href="mailto:Asterisk@muug.mb.ca">Asterisk@muug.mb.ca
</a><br><a href="http://www.muug.mb.ca/mailman/listinfo/asterisk">http://www.muug.mb.ca/mailman/listinfo/asterisk</a><br><br></blockquote></div><br><br clear="all"><br>-- <br>Sean Walberg <<a href="mailto:sean@ertw.com">
sean@ertw.com</a>> <a href="http://ertw.com/">http://ertw.com/</a>