> Now multiply that by thousands of concurrent connections each doing multiple I/O operations. Servers spent ~95% of their time waiting for I/O operations.

Well, no. The particular thread of execution might have been spending 95% of time waiting for I/O, but a server (the machine serving the thousands of connections) would easily run at 70%-80% of CPU utilization (because above that, tail latency starts to suffer badly). If your server had 5% CPU utilization under full load, you were not running enough parallel processes, or did not install enough RAM to do so.

Well, it's a technicality, but the post is devoted to technicalities, and such small blunders erode the trust to the rest of the post. (I'm saying this as a fan of Bun.)

If you’ve never seen a machine stuck waiting for disk I/O, I don’t know what to tell you… but it is common, even with SSDs it can happen (as they point out due to handoff between the OS and user level process takes time)