[CentOS] Load Balancing

Tue May 23 19:46:48 UTC 2006
Fabian Arrotin <fabian.arrotin at arrfab.net>

On Tue, 2006-05-23 at 10:38 -0700, Dan Trainor wrote:
> Chris Mauritz wrote:
> 
> > Mace Eliason wrote:
> >
> >> Hi,
> >>
> >> We are starting a new project, and are trying to decide the best way 
> >> to proceed.  We want to setup a LAMP configuration using Centos, 
> >> something we have been doing in the past with great success.
> >>
> >> The question is load balancing.  We antisipate the potential for the 
> >> system to receive 500,000 requests/ day with in the next year.  We 
> >> want to plan for that extra load now as we start the project.  What 
> >> would you suggest for setups for multiple servers for redundancy and 
> >> load balancing?
> >>
> >> I have setup MySQL replication and that works fine but what about the 
> >> rest of the system.  I know it is quite simple to setup with windows 
> >> 2003 server.
> >>
> >> Would a cluster be the way to go?  Ideally we would like 2-? severs 
> >> setup that are all identical and sharing the load as need be, and if 
> >> one fails users would notice nothing.
> >>
> >> I have also thought of just looking for a hosting company that offers 
> >> load balancing servers and not worry about it but we like to have 
> >> control.
> >>
> >> Thanks for any suggestions
> >
> >
> > Perhaps this will help get you started:
> >
> > http://www.howtoforge.com/high_availability_loadbalanced_apache_cluster
> >
> > Cheers,
> 
> Hi -
> 
> FWIW, I've been toying around with this as well.  Right now I'm trying 
> to decide which shared storage mechanism we'll be using for the nodes 
> themselves.  We need to keep the data consistant across 10+ machines, 
> which will be serving this content.
> 
> If this hasn't been mentioned before, I've been using LVS for a while, 
> with a whole lot of success.  It's smart, scalable, and works quite 
> well.  If you're looking for an open-source load balancing and 
> distribution system, I highly suggest you investigate this.
> 
> If anyone wouldn't mind chiming in with some ideas, I'd greatly 
> appreciate it.  I'm sure others would, too,
> 
> Thanks!-
> -dant
> _

For the backend storage, it depends what's your budget ... :o)
A minimal setup is to use nfs on a central server to host/share the same
data across all your machines ... the problem in this config is that the
nfs server becomes the single point of failure ... so why not using a
simple heartbeat solution for 2 nfs servers acting as one and uses drdb
between these 2 nodes for the replication ...
Other method is to have a dedicate san with hba in each webservers but
that's another budget ... :o)

Just my two cents ... 

-- 
Fabian Arrotin <fabian.arrotin at arrfab.net>