[Gllug] Geographcial server failover

[Tethys]

Darren Beale writes:

>Any sort of scripting or round robin DNS does not sound like it'll cut
>it, there is going to be some lost data that way.

Yep. But who cares? The costs of that level of availability are very
high. As you approach 100% uptime, the costs rise exponentially. Very
few businesses genuinely need the availability so badly that they're
prepared to pay what it costs. Most will decide they can live with a
small level of outage when shown the prices for preventing that outage.
Data General used to have a paper explaining the issues, and would
guarantee up to 6 nines availability for their systems if you were
prepared to pay the money for it (IIRC, that was 7 figures, and that
was only for single site systems).

>What I'm looking for is a way for requests to foo.bar.com to switch
>from one machine in a datacentre in - say - Redbus Docklands, to
>another identical server in a Glos or Manchester centre if the first
>one goes down.

Despite what I said above, there are cases, when the company *does*
need the availability. There are two common approaches to doing this:

1. Do it through DNS.
2. Do it through BGP.

This paper tell you all you need to know about the available options
(and why you may want to consider not doing it at all):

	http://www.tenereillo.com/GSLBPageOfShame.htm

We're going the BGP route (we can't do it through DNS, as we're not
authoritative for some of the web sites we host, which is a prerequisite
for using DNS liek this). We have a point to point link between our sites,
which helps mitigate some of the BGP based problems (routes changing mid
sessions, for example).

Tet
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug