load balanced configuration

[Alex Dean]

On May 20, 2010, at 5:31 PM, Bryan O'Neal wrote:

> Personally I vote for RRDNS so that your domain name has multiple IP's
> associated with it. DynDNS polls every few minuets for availability
> and will automatically remove dead servers. That is what clustering is
> all about :)

Keith originally mentioned wanting servers in 2 distinct locations, a  
primary site and a backup site, with the backup site able to take over  
automatically if the primary becomes unavailable.  To me, this sounds  
like a concern for availability, not performance, and my comments have  
been made in that frame of mind.

'clustering' can mean a lot of things.  Increasing performance, as in  
high-performance computing, is not the same as increasing  
availability.  The kind of load-balancing you can achieve through  
RRDNS does not necessarily increase availability.  You have to  
consider how well your current infrastructure is matched to your  
current workload.

What I mean is this : If you have 2 servers doing an identical job  
(like 2 web servers serving up the same website, to the same users,  
etc), you can load-balance through RRDNS, or through a dedicated load- 
balancer, or whatever.  That doesn't automatically do anything to  
increase your site's availability.  If your site's traffic load  
requires both servers to be functional in order to get decent  
performance for the user, then losing 1 server means your site is  
effectively unavailable.  By contrast, if the 2 servers are in an  
active/passive configuration, you aren't load-balancing at all, but  
you do have high-availability.  If the primary server dies, the old  
secondary can become primary and users should never know the difference.

I think it's really important to keep these goals distinct, assign  
relative importances to each, and act accordingly.  High performance  
is not high availability.  High availability is not backup.  Decide  
which you need.  Hint: You ALWAYS need backup. :)

I often hear people bemoan the fact that in a typical HA setup,  
they've spent money on all this backup hardware, and it's not 'doing  
anything'.  It's just sitting there waiting for the primary server to  
fail.  Yes, this is true.  You just have to keep in mind that if it  
were 'doing something', you'd miss it if it died, and that's not  
really high-availability.

alex