160 memory modules of 8GB each. That means 1.3 terabytes, or 1 300 000 megabytes of RAM. Numbers that put a grin on most of our faces. Our beloved CEO, Sylvain Leclair said: “If you had told me, when we installed our first shared hosting server with 256MB of RAM in 1996, that we’d be adding over a million megs of it in our servers one day, I would not have believed it!” Even if Sylvain is fascinated by these impressive numbers, our technicians already inhabit the language of the terabyte in their daily lives. This does not mean that 1.3TB of memory is not a considerable amount or something to brush off. This isn’t disk space!
Memory usage on our servers averages below 50% of their total capacity, allowing us to provide optimal performances at all times. Therefore, why add more memory? We prefer speaking of prevention, because all those RAM modules will mostly be used for server redundancy.
New Technologies, Resources & Performances
This considerable amount of memory will offer faster loading time, and less risks of server resource saturation. Additionally, increasing memory allows us to support new technologies that can be resource hungry.
For example, we have recently replaced the module managing PHP requests to FastCGI. Even though FastCGI is more efficient and processes requests faster than suPHP (the prior version we were using), it is also more resource hungry. We knew that FastCGI required more server memory, but did not expect it to be so significant.
Some may say that this increase in server resource consumption, that can sometimes go near the 60% mark, is negligible, and leaves plenty of buffer space to absorb unexpected events. They wouldn’t be wrong! However, if we observe our quality standards and the resources required for our redundant platform, this situation is unviable.
Safety Net For Redundancy
Our hosting servers are set up on a high quality redundant infrastructure. In short, Server A replicates in real-time on Server B. In the eventuality something breaks on Server A, Server B takes over and takes all the traffic and load, making sure that Server A’s services are uninterrupted. In that eventuality, we need twice as much memory to ensure that server performances remain steady, because two servers have now merged into one.
The switchover from one server to another is something that does not come up often (a few times a year), does not last very long (less than 24 hours), and normally causes very little slow downs during the intervention period. However, we consider that negative impacts during an intervention should be limited at all costs, and ideally not even be perceived.
All in all, we are adding 1.3TB of RAM on our servers in the upcoming weeks, for the benefit of all!
