For the technically minded, we went from a t3a.small EC2 instance to an m6a.large. Both are dual-core instances, but the t3a.small type is on AMD’s Zen 1 architecture, while m6a.large is Zen 3 based, which is light years ahead (this is the equivalent of swapping out a Ryzen 1600 to a 5600X). On top of that, on the T instances you’re expected to only use 20% of the CPU and there’s a pretty heavy surcharge if you continuously use more, while on the M instances you just get the whole thing, so hosting costs are actually projected to be lower even with the jump in CPU capacity, plus we enjoy a nice jump from 2 GB of memory to 8 GB.

And yeah, while the conventional wisdom is that burstable instances are a great fit for web services, we actually have a remarkably flat CPU load. I’m thinking it’s probably federation that drives most of it it rather than direct usage of the site. I’m also fairly sure a c6a.large instance would be more cost-effective (it’s basically the same thing with half the memory for 12% cheaper) but that’s something to maybe figure out later.