"Don't replace platform/native threads with virtual ones, replace tasks (without further explanation) instead"?!
Combine that with the fact that they chose to implement the scheduler in Java instead of C(++) and you're set for performance problems.
Remember that NIO took from 1.5 to 1.7 to be usable/performant and that was native!
Edit: Finally figured out why: https://news.ycombinator.com/item?id=39010648
The JDK has historically used some native implementations in its stdlib (zip, imageio and others), back when the runtime wasn't as fast as it is today. But today's runtime would often be faster in Java than those native implementations.
Ah yes, the argument from the 1990s. It would make sense to understand where the JVM and its compiler are these days before making incorrect statements about performance.
From your link:
> Blocking network TCP IO needs a sychronized block to work
This is utterly false.
I have always had to do synchronized(something) { socketInputStream.read(); }
And the dude himself says that reading from a socket is a problem if you listen to the interview.
Synchronized in this context is pretty nonsensical.