Messages in this thread | | | Date | Thu, 29 Jul 2004 10:23:05 -0700 | From | William Lee Irwin III <> | Subject | Re: [PATCH 20/22] AIO poll |
| |
On Thu, 2004-07-29 at 18:02, Avi Kivity wrote: >> I second the motion. I don't see how one can write a server which uses >> both networking and block aio without aio poll.
On Thu, Jul 29, 2004 at 06:16:07PM +0200, Arjan van de Ven wrote: > one could try to use epoll and fix it to be usable for disk io too ;)
That notion doesn't make any sense. epoll is just a continuation-based implementation of a recurring readiness notification operation. Extant async I/O operations (in mainline) are data transfer. The inclusion goes the other direction. Readiness notification can be modeled as a kind of data transfer by synthesizing a data stream to represent the readiness transitions, but the reverse fails as data can't become ready to consume (what epoll would like to report) until I/O is initiated. epoll is suited to become a recurring aio readiness notification operation. Vice-versa is literally so logically inconsistent it can't even be phrased properly.
I'll send in some code to deal with this properly later on today. New polling infrastructures should not be erected, but rather existing epoll code reused for continuation-based readiness notification.
-- wli - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |