On 9/12/07, <b class="gmail_sendername">Thomas Ptacek</b> <<a href="mailto:tqbf@matasano.com">tqbf@matasano.com</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Doesn't POSIX aio give you async filesystem access? That would save<br>most of my designs at least one thread.</blockquote><div><br><br>Have you really found that file i/o is slow enough to be worth doing on a separate thread?
<br><br>If you look carefully in the EM code, you'll find the traces of the time I started adding event-driven file i/o. I never finished it because it didn't seem worth the trouble, but maybe I was wrong.<br><br>
There is a lot of things in the file i/o channel that change perceived performance, like the firmware on the disk controller, the behavior of your SAN card if you have one, whether you use hardware or software RAID, and of course the buffering that the kernel itself does. Is there anything unusual about your systems or your software that makes disk i/o particularly latent?
<br></div><br></div>