Re: redundancy via DNS

[Craig Sanders <cas@taz.net.au>]

On Wed, Jun 20, 2001 at 10:49:24AM +0200, Marcel Hicking wrote:
> Maybe the Linux Open Router project could be helpfull. It is in a very
> early stage but AFAIK it aims at similar tasks.
>
> http://www.open-router.org/about.html

linux virtual server (LVS) is a good linux-based load-balancer (layer
4 switch, not DNS round-robin). 

i've used LVS for large squid proxy-cache arrays and am about to use it
for a large array of web servers. it's remarkably easy to set up, and
does an excellent job. highly recommended.

more details on LVS can be found at:

http://www.linuxvirtualserver.org/


LVS supports load-balancing of servers at different physical locations,
but it's extremely difficult to eliminate all single-points-of-failure -
most of the work is going to be in *designing* your redundant network,
rather than implementing it.

e.g. as a starting point, think about DNS round-robin with A records
pointing to two or more LVS load-balancer boxes at different locations,
using Direct or Tunnel methods to get to the individual real-servers.

you'd need several secondary NS machines, low TTLs on the DNS records
(which won't defeat all caching - some admins ignore TTL), and tools
to change the DNS records if one of the network paths went down...and
then push the changes out to the secondary NS machines. this is, of
course, a single-point-of-failure in itself....if your primary NS goes
down or is unreachable for any reason then everything is affected. so
you'd need a second one located somewhere else which can take over if
necessary...complicating the setup even further.


don't be discouraged by these comments, load-balancing (whether with LVS
or another tool) is still very useful but it's not a magic solution to
all outages.



craig

-- 
craig sanders <cas@taz.net.au>

Fabricati Diem, PVNC.
 -- motto of the Ankh-Morpork City Watch