Thank you for your patience.
MailRoute has taken significant steps to manage our growth by having a presence in two data centers, and we are opening a third data center today while significantly altering our architecture to provide another level of redundancy and to increase the reliability of our entire system.
In our old architecture, the balancing of traffic amongst datacenters and the servers inside of them had shared component - the routers and loadbalancers exchanged regular information about network and server health to determine the optimal route and destination for all traffic. The idea is that the low-level network statistics (speed, location, number of connections, etc) were adequate to properly load-balance traffic amongst all our servers in our datacenters. MailRoute’s configuration has included, for many years, a significant investment in Cisco and Foundry hardware to implement this model.
In the past 72 hours, we have seen issues with that hardware, where an issue in the Cisco routers is affecting the Foundry Load Balancers. This is not how the hardware is meant to operate, but it is happening. Neither company seems to be able to offer a reliable fix.
Fortunately, we were simply a few days away from moving to a new architecture. We have been running a subset of our service on this new system for the last 6 months, and the latest trouble is making it clear that it's time for us to do the final migration.
We’re moving off of Cisco routers onto Juniper. We are moving off Foundry load balancers to custom software-based load balancers that can better evaluate the email characteristics of our unique environment.
Yes, we're doing this a little ahead of schedule - but not by much, only by a few days. This was scheduled to be happening this weekend, and instead we've started rolling it out now. In short, we are moving away from a significant investment in our existing hardware to create a permanent fix to this problem.
The change should be completely transparent to all our customers, except that you should see improved speed and connectivity and any downtime issues should go away. Inbound and outbound traffic, instead of sharing a central component, will all be completely independent in each datacenter.
Thomas A. Johnson
Please sign in to leave a comment.