Anycast...any one using it on IP4

Barry Margolin barmar at alum.mit.edu
Sat Aug 21 17:10:20 UTC 2004


In article <cg7rs1$1a28$1 at sf1.isc.org>, Paul Vixie <vixie at sa.vix.com> 
wrote:

> Barry Margolin <barmar at alum.mit.edu> writes:
> 
> > We were using it in a big way when I was at Genuity.  ...
> > 
> > We later expanded it to our authoritative servers as well.  There are at
> > least two each of DNSAUTH1.SYS.GTEI.NET, DNSAUTH2.SYS.GTEI.NET, and
> > DNSAUTH3.SYS.GTEI.NET.  So there's multiple levels of redundancy and load
> > sharing -- the stuff performed inherently by the DNS protocols, as well
> > as that provided by the anycasting.
> 
> i have two questions.
> 
> 1. was that before or after the acquisition of "bechtel's genuity" and
>    their load-balanced (noncoherent) anycast?

It was after GTE acquired BBN Planet, which they did at the same time as 
they acquired the old Genuity, because all the anycasting servers' names 
are in the GTEI.NET domain.

> 2. is "gte's genuity"'s authoritative dns anycast system load-balanced
>    (noncoherent) or is it just for load-sharing/redundancy these days?

Unless Level(3) has made significant design changes since I left in 
December, it just uses OSPF's routing table to make the decision.  So 
it's just redundancy/load-sharing, because it doesn't have any way of 
testing server loads.

Of course, if a particular server gets loaded enough to make a 
noticeable impact in its response time, the client nameservers will 
switch to one of the other IP's based on their response time algorithm.  
So DNS will automatically load balance to some extent.  Without anycast, 
network latency is likely to be the major component of the response 
time, and will often be the deciding factor with geographically disperse 
nameservers.  But with anycasting, we can have complete sets of servers 
on each coast, so the server capacities and loads will have more of an 
impact.

-- 
Barry Margolin, barmar at alum.mit.edu
Arlington, MA
*** PLEASE post questions in newsgroups, not directly to me ***


More information about the bind-users mailing list