Date: Fri, 22 Jun 2001 18:53:29 -0700 (PDT) From: Richard Hodges <rh@matriplex.com> To: Josh Osborne <stripes@mac.com> Cc: freebsd-hackers@FreeBSD.ORG Subject: Re: question: aio / nbio / kqueue Message-ID: <Pine.BSF.4.10.10106221838150.65610-100000@mail.matriplex.com> In-Reply-To: <200106230044.RAA18157@smtpout.mac.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, 22 Jun 2001, Josh Osborne wrote:
> [...]
> >> AIO is good when you are not receiving much data (or not receiving
> >> it very frequently), and presumably want very low latency.
> >
> > What if you want good performance with "moderate" disk IO, say ten
> > to twenty megabytes per second continuously?
> > I tried AIO some months ago (4.1R or 4.2R), but had some trouble
> > with AIO, mainly that it seemed to lose track of half my files.
> > Not any particular files, it seemed that at any moment it would
> > just pick ten or so (out of maybe 20-25 files) to ignore at any
> > given time.
> I try to avoid anything that makes me write signal handlers (AIO
> is done with signals, right?), and doing a ton of data block after
> block doesn't sound like AIO's bag anyway. Maybe you missed the
> AIO signals by failing to properly do some mystical signal ritual.
That could be, but I do *try* to be aware of these things :-) My
main goal is to keep a decent read backlog on the drives at all times.
Or looking at a differnt way, I want to minimize the time that each
drive is idle.
> > Is there any better solution than just forking off a process for
> > each file and letting the kernel handle the details?
> Depends, is kernel hacking strictly off limits?
Not at all. As a matter of fact, I have been writing (and rewriting)
quite a bit of kernel networking code. Most is reserved for a product
for sale, but I can share a few bits, like the ForeLE/IDT ATM driver.
> If you do have to fork child processes, using sendfile (unportable)
> may well have pretty low overhead. It is documented as zero copy.
> I would give that a shot.
Sendfile is TCP, isn't it? I thought so. I am sending UDP, and the
timing needs to be *perfect*. Well, within a couple milliseconds,
anyway.
Thanks for the info,
-Richard
-------------------------------------------
Richard Hodges | Matriplex, inc.
Product Manager | 769 Basque Way
rh@matriplex.com | Carson City, NV 89706
775-886-6477 | www.matriplex.com
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.10.10106221838150.65610-100000>