Kevent take 26

Posted Dec 14, 2006 16:05 UTC (Thu) by kjp (guest, #39639)
Parent article: Kevent take 26

After casually looking at a few iterations of this but trying to figure it out, I hate ths secondary ring buffer API. It's fugly, I can't tell from the writeup if it actually saves any copies, and now it takes *two* syscalls per loop (a commit and a wait) to process more data. Ugggh!

It reeks of premature, wanna-be optimization.

I really hope they just go with the synchronous API and let non-trivial apps people beat on it a while.

to post comments

Kevent take 26

Posted Dec 15, 2006 18:17 UTC (Fri) by nevyn (guest, #33129) [Link]

I completely agree, it's not like doing one syscall and copying the events is the big speed problem. Ulrich seems to be pushing it to be even more fugly just so it works "well" with multiple threads all competing for the same events ... yeh, like I'd trust anyone to get the data access right if they are doing that.

But whatever, if the kernel guys want to include a big POS for slow unstable apps. I'll happily just use the sane API with multiple processes.