“About 40 minutes into Hour 6 your instance no longer has any credits remaining. At this point, you’re now limited to the baseline performance of the instance, in this case, 10% of the vCPU.
At this point your application grinds to a halt.”
Agreed. As terrible as that is, it’s actually worse than that in my experience, bc even while you have positive CPU credits, you can’t spend them down smoothly to achieve high sustained CPU - the scheduler gives you something more like a sine wave of performance. Which is mystifying to say the least and leads me to my final point about why T2 is so awful: it violates the principle of least astonishment. Why would anyone expect that a CPU is limited to a small fraction of its output? T2 should not be categorized in the “general computing” aws instance family. It should be in a family all its own called “burstable” that explains the major difference between it and the general computing family.