[128845] in North American Network Operators' Group
Re: Numbering nameservers and resolvers
daemon@ATHENA.MIT.EDU (Robert E. Seastrom)
Wed Aug 18 09:37:31 2010
To: Mikael Abrahamsson <swmike@swm.pp.se>
From: "Robert E. Seastrom" <rs@seastrom.com>
Date: Wed, 18 Aug 2010 09:37:09 -0400
In-Reply-To: <alpine.DEB.1.10.1008172134060.8562@uplift.swm.pp.se> (Mikael
Abrahamsson's message of "Tue, 17 Aug 2010 21:34:58 +0200 (CEST)")
Cc: nanog@nanog.org, Nick Olsen <nick@brevardwireless.com>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
Mikael Abrahamsson <swmike@swm.pp.se> writes:
> On Tue, 17 Aug 2010, Nick Olsen wrote:
>
>> So lets say that you have multiple DNS resolvers in the same ip space that
>> you advertise from multiple locations. All would be fine for the most part.
>> But if you had a location equidistant network wise from two POP's wouldn't
>> it load balance and possibly break some TCP sessions? How would someone get
>> around this? This is also what OpenDNS does from what I understand.
>
> Usually network do not loadshare per-packet on BGP, so a TCP session
> will "always" go to the same dns server, at least for the short
> duration this TCP session lives.
Occasionally I have seen networks (usually small dual-homed ones) that
attempt to equally utilize their network pipes by doing per-packet bgp
load balancing to both upstreams. Then they wonder why their
performance is so irregular. :-)
-r
