This is the fifth Stack Decision of this series. Read the last one to catch up.
One thing that happens when your service gets really popular is that you start running into server capacity issues. From day 1 we launched with #SSL, which might seem like “well duh” today, but it wasn’t that common in 2012. Jump to circa 2015 and the service is doing ~4k request per second (on average) and over 75% of those are SSL, puts a lot of stress on the CPUs.
At the time, MaxCDN deployed a custom build of nginx and we optimized the hell out of it. What we ended up having to do was basically dedicating a server in the busiest datacenters (PoPs) just for BootstrapCDN. Most enterprise customers didn’t even get their own (because they didn't need it). The engineering team did some other cool optimizations but I honestly can’t remember verbatim. Hopefully, they can comment on this. I’m lucky because this allowed me to work with some really great engineers, inside and outside of MaxCDN. Specifically an engineer by the name of XhmikosR. That’s in part 6.
AMA below. 👇