Weekly edition	Kernel	Security	Distributions	Contact Us	Search
Archives	Calendar	Subscribe	Write for LWN	LWN.net FAQ	Sponsors

LCA: A new approach to asynchronous I/O

Posted Jan 29, 2009 3:06 UTC (Thu) by kjp (subscriber, #39639)
Parent article: LCA: A new approach to asynchronous I/O

I honestly fail to see a need for this. In fact I bristle at the thought of this or any other "we're too cool for syscalls so lets have yet another event ring" going in. And Yet Another Wait mechanism that doesn't look epoll compatible.

I'd rather manage my own thread pools in user space than figuring out this crazy api which essentially does the same thing.

(Log in to post comments)

Posted Feb 10, 2009 21:03 UTC (Tue) by ddaa (guest, #5338) [Link]

Seconded.

According to the mailinator blog, blocking I/O wrapped in threads using NPTL smokes non-blocking I/O in terms of performance. It is also fairly straightforward to implement at the application level.

There are certainly reasons to implement the POSIX standard, mostly political I guess. But for pragmatic application development on operating systems that have good threading, non-blocking I/O seems like more trouble to use than it's worth.