New Heluna Architecture

We recently migrated our architecture to a new, faster cluster, and we thought we’d mention just what was required to make the transition.

First: get the operating systems online, and hardened for security. When you’re dealing with spam, virus writers, etc, you simply can’t be too careful. Also, as we pass confidential data (e-mail messages) for our clients from system to system, we maintain as much security as possible at every level, from the operating system all the way up to the custom software we’ve written.

Second: get the core software running. The new cluster runs a newer version of the MySQL database, in a clustered environment, so it’s faster and it’s more reliable than our previous database. MySQL is notoriously hard to upgrade in-line, which is why we waited until cluster migration in order to upgrade. Also, we have fresh copies of our mail servers, the software that bridges the gap between our antispam filters and the mail servers, and other various bits and pieces that make up all of our core software.

Third: redesign and redeploy our message routing system on the new cluster. With the extra speed of new hardware, combined with a few ideas that we had running in our test environment, we’re able to make sure that messages get filtered even faster.

Fourth: get the website running on the new cluster. This was actually our easiest and fastest step; while www.heluna.com was running on our old cluster, we had a parallel website running on the new cluster, with full functionality, ready for clients to be deployed.

Fifth: extensive testing. The heluna.com domain ran on the new cluster for a full quarter prior to rolling it out to any of our clients. Upon being completely satisfied with the performance of the new cluster, we prepared for launch.

Sixth: launch! mail.heluna.com was switched over in late May, so all incoming e-mail messages are now going to our new cluster. We’re happy to report that for nearly all of our clients, the transition was seamless. A few of our clients, due to various custom configurations, are still getting their messages routed to our old cluster, but we’re moving more clients over every day, and we expect all clients to be running on the new cluster soon.

As you may expect, transitioning a system that, by its very nature, must be much more reliable than, say, Twitter, is rather difficult, but I’m happy to say that the Heluna transition has been extremely successful.