Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 5 Aug 2003 07:42:21 +0200 (CEST)
From:      Blaz Zupan <blaz@si.FreeBSD.org>
To:        "David J. Hughes" <bambi@Hughes.com.au>
Cc:        freebsd-isp@freebsd.org
Subject:   Re: DNS Server Farm
Message-ID:  <20030805073757.G78439@titanic.medinet.si>
In-Reply-To: <20030804152939.F97726@elk.hughes.com.au>
References:  <20030804182222.A5C055553E@titanic.medinet.si> <20030804152939.F97726@elk.hughes.com.au>

next in thread | previous in thread | raw e-mail | index | archive | help
> An anycast / BGP solution assumes that the application will never fail
> without taking the box down.  If your DNS code falls over (or is stopped
> etc) then your customers are going to lose.

Running a watchdog that checks if the DNS code is alive is next to trivial. If
the DNS server fails, either restart it or take down the routing protocol so
the system fails over to the next anycast box. If the machine stops to the
point that the watchdog is also not working anymore, then the routing protocol
doesn't work anymore as well, so you fail over to the next anycast box.

> Seeing as a "real" load balancer, like a ServerIron XL, can be found on
> Ebay for next to nothing these days, it's a very workable solution.  We
> just built such a solution for a network with about 550,000 users and it
> works just fine.  2 or 3 good boxes and a load balancer will also be much
> cheaper than a box at each pop.

But you still have a single point of failure. You have all your boxes located
at a single location. If you have a power failure or other catastropic event
at that location, your whole DNS setup fails, while the rest of the network
works just fine, just without a working DNS server. Same thing if your load
balancer fails.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030805073757.G78439>