[ruby-core:86915] Re: [Ruby trunk Feature#13618] [PATCH] auto fiber schedule for rb_wait_for_single_fd and rb_waitpid

[Eric Wong <normalperson@yhbt.net>]

samuel@oriontransfer.net wrote:
> I've been thinking about these issues for the past few days, and one thing I found was this video https://www.youtube.com/watch?v=KXuZi9aeGTw#t=519 where they specifically talk about the latencies incurred by the threading model you propose. I just thought you might find it interesting.

Can you please summarize and perhaps link to code they talk about?

I don't have video now and I don't want it to become an expectation
that Ruby contributors be able to deal with video
(either through hardware, software or human physical limitation)


No rush, either, barely have any computer access for a bit.


Again, Mio (Glasgow Haskell Compiler) is most similar to what
I'm working on, here, and that scales to some ridiculous number
of cores:

http://haskell.cs.yale.edu/wp-content/uploads/2013/08/hask035-voellmy.pdf

I don't use a dedicated epoll thread like they do;
but I also know epoll_wait uses exclusive wakeup in Linux,
so there's no thundering herd when a single event arrives
when multiple threads are sleeping in epoll_wait.