Minimal Failsafe Configuration Options

[worthstream]

Hi,

we're trying to use OpenX (for the first time in the big company where i'm working) to deliver some campaigns on third parties sites.
Everything has been implemented via direct ad selection, and is working really well on the single machine test, so we're ready to move to a distributed production environment.

This is where i need some help: we're not going to deliver too many impressions (in the order of ten millions per month) but having no single point of failure is absolutely necessary.

I was thinking of using two frontends behind a load balancer that will redirect all of the traffic to the working machine, if the other fails.
As for the backend, i was leaning toward two machines with db mirroring, so that we can do a cold swap in the event of a failure of either.

I'd really appreciate any comment on this configuration, and any input on the hardware that those machines should be equipped with.

Thank you very much!

[Arlen Coupland]

Hello,

A single server for OpenX should easily get you above 50m/month, and the cealing will depend on many factors (not just hardware)

If this sounds fine, then using 1 server as a pure backup (if you have such resources) is always a good idea. Simply many people don't have such hardware to dedicate to this.

Another option to consider as it sounds like you have 4 machines, is the OpenX Distributed Statistics method. In this setup there is 1 central machine and multiple delivery slaves. The slaves contain a replicated version of the database and are under a load-balanced delivery URL. Every 15mins they copy their stats to the central machine which will summarize + store all statistics.
With this method - if 1 delivery slave fails, the others can still be used. If central server fails, delivery servers continue and simply must queue their stats until the central server is back.

Some links:
http://blog.openx.org/10/serving-billions-...ds-using-openx/
http://www.openx.org/docs/2.4/whitepapers/...uted-statistics
http://www.openx.org/en/docs/whitepapers/performance-tuning