>On Fri, Jul 06, 2001 at 01:37:42PM -0700, glen mccready wrote:
>> anyway, there's more to doing posix aio properly than immediates jumps
>> to mind and that's what i was pointing out... you also don't need a
>> thread per process (or per aio event) to put together a "reasonable"
>> hack job. :-)
>
>Of course, there's also (or was, once, briefly) a "BSD aio" abstraction,
>and if someone were going to pour the work into actually doing aio "right"
>it might at least be worth a look -- see Fall and Pasquale on splice().

http://citeseer.nj.nec.com/rd/94786292%2C9849%2C1%2C0%2CDownload/http%253A%252F%252Fwww-cse.ucsd.edu/groups/csl/pubs/conf/usenix93.ps

for anybody interested.  i'm not sure this qualifies as the "right" way to
do aio, but it would certainly help the case of accessing a raw disk... it
doesn't really touch upon dealing with a filesystem, and it only briefly
touches upon joining two udp sockets, but makes no mention of tcp.  ubc
also makes this a tougher proposition since the pages just get faulted in
rather than brought in via bread() or breadn()...

is that the paper you were thinking of, thor?

glen