[Linux-HA] HA with Commodity rented servers

[Jonas Andradas]

Hello Ed,

I am most interested in the scenario you are talking about, so I will follow
this thread anxiously, and will be very glad in anybody shines some lite on
it, and gives any ideas :)

Regards,

Jonas Andradas

On Tue, Feb 12, 2008 at 8:26 PM, Ed W <lists at wildgooses.com> wrote:

> OK, how feasible is it to rent some commodity servers from the likes of
> a cheapo host like 1and1 and turn them into something robust?
>
> Right now I have had some success taking a rented box and using vservers
> to run multiple virtual machines.  It's then easy to add new physical
> servers and shunt services around.  However, we had the hardware lockup
> on one machine recently and suddenly everything on that machine becomes
> unavailable...  So now I'm trying to figure out how to make this system
> robust against hardware failure as well.
>
> Apps are a mailserver (Dovecot+Postfix) and a custom Rails webapp
> (currently under Apache, but considering alternatives).
>
> Clearly it's a piece of cake to build two identical mail/webservers, the
> bits I am not quite understanding are:
>
> 1) How to load balance to my hypothetical backend of 3 (or so) identical
> machines?  (remember I am using some rented boxes so I can't shift IP
> addresses around and have just 100mbit ethernet between everything)
> 2) How to create reliable persistant storage for the mail and webservers?
>
> Assuming only the rented boxes described (as many as you like!), is it
> possible to do better than simply playing DNS tricks to keep
> availability if backend machines go down? (ie solve (1) above).  Seems
> like I can either use simple dns round robin to the backend servers if
> thats all I need, or I can fit two loadbalancer machines (or some
> description) in front if I need better balancing - however, one way or
> another I'm back to the situation that the loadbalancer hardware could
> go bad?
>
> There is conflicting advice on best practice to solve 2) above.  Given
> only a single net connection to each of the machines, split brain type
> problems seem like a normal situation?  However, this particular host
> provides a net based serial console and also a net based method to
> forceably reboot the machine (or take it down), however the later is not
> instant and may take some fiddling to automate usage of these features.
> Also the limitation is still that net interface... What happens when
> someone blows up an internal router and the network becomes subtly
> internally broken?
>
> What would be best for this situation to run reliable mysql + mail
> storage + webserver storage?  Gluster?  DRBD + OCFS ?? Something else?
> (Pay through the nose for a san...?)
>
> Grateful for some pointers to get started here?
>
> Ed W
>
>
>
> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>



-- 
Jonás Andradas

Skype: jontux
LinkedIn: http://www.linkedin.com/in/andradas
GPG Fingerprint: 5A90 3319 48BC E0DC 17D9
130B B5E2 9AFD 7649 30D5
Keyservers: wwwkeys.pgp.net | pgp.rediris.es