[matachuan]

lol I'm well informed on this topic but thanks anyway. Maybe I didn't make myself clear in my previous post. I was questioning the motivation -- Is there any existing profiling work that shows a significant amount of time has been spent on context switches in this particular workloads and how much you can save by adopting this approach.

[eptcyka]

Well, do you not recall the massive amounts of moaning that took place because now everyone's I/O bound stuff (most webservers/databases) works far slower than it used to when spectre/meltdown patches had been applied? If we could get rid of the cache flush that occurs when a userspace process syscalls, there'd be a lot less heat produced at your local data center.

[matachuan]

I agree with what you said because everyone knows it's massive. All I wonder is how massive, quantitatively.

[candiodari]

Unikernels are generally considered especially advantageous in network servers.