One thing that spoiled me at Pardot was using Softlayer as our hosting providing – a great thing that they do is have a private fiber line between each of their data centers. Setting up a possible contingency plan on Amazon OpsWorks, I found this wasn’t the case.
David and Sthithia set up Pardot so that in case of “nuclear strike” – or the more likely short blip where an entire datacenter goes offline, no Pardot data would be lost, and as importantly, this was a live replication that could be tested at any moment.
Setting up a second server on Amazon (our RDS was in N. Virginia) in Oregon for this geo-redundancy showed me the powers of traditional hosts and something Amazon needs to improve upon. Our tester was complaining about slowness on Oregon, confusing since it was a production environment and rails with the same m3.medium size server.
Anyway, once I diagnosed the issue and decided that changing the OpsWorks cookbooks for the slave were going to be much more trouble than just moving to a different AZ in Virginia.
What do you recommend? I figure that Heroku can’t handle this any better. Makes me miss Softlayer and Rackspace.
Here’s some marketing.