Re: async I/O seems to be blocking on 2.6.15

[Brent Baccala <cosine@freesoft.org>]

On Fri, 3 Nov 2006, Jens Axboe wrote:

> On Fri, Nov 03 2006, Brent Baccala wrote:
> > ...
> >
> > The enqueues still take a noticable amount of time, though, just a lot
> > less than before.  They average 1 second total.  That's 100 one-MB
> > reads, broken down into 128 KB blocks, I suppose, for a total of 800
> > low-level reads.  Setting nr_requests higher (2048) doesn't seem to do
> > any more good.
> >
> > I can see that you've put me on the right track, but I am still
> > puzzling... any idea what the remaining second is being used for?
>
> Try to time it (visual output of the app is not very telling, and it's
> buffered) and then apply some profiling.

Yeah, I did time it.  I bracketed the enqueues with calls to
gettimeofday().  It takes them about a second to run, and I made sure
that the fprintfs were outside the timing checks.  Here's the actual code:


    fprintf(stderr, "Enqueues starting\n");
    gettimeofday(&tv1, NULL);

    for (i=0; i<NUMAIOS; i++) {

        memset(&iocb[i], 0, sizeof(struct iocb));

        iocb[i].aio_lio_opcode = IOCB_CMD_PREAD;
        iocb[i].aio_fildes = fd;
        iocb[i].aio_buf = (unsigned long) buffer[i];
        iocb[i].aio_nbytes = BUFFER_BYTES;
        iocb[i].aio_offset = BUFFER_BYTES * i;
        /* aiocb[i].aio_offset = 0; */

        iocbp[0] = &iocb[i];
        if (io_submit(aio_default_context, 1, iocbp) != 1) {
            perror("");
            fprintf(stderr, "Can't enqueue aio_read %d\n", i);
        }
    }

    gettimeofday(&tv2, NULL);
    subtract_timeval(&tv2, &tv1);
    sprint_timeval(strbuf, &tv2);
    fprintf(stderr, "Enqueues complete in %s\n", strbuf);


And the output:


baccala@debian ~/src/endgame$ time ./testaio
Enqueues starting
Enqueues complete in 1.187s

real    0m5.335s
user    0m0.001s
sys     0m0.760s
baccala@debian ~/src/endgame$


What kind of profiling did you have in mind?  Kernel profiling?  As
you can see, its user time is basically nil.



					-bwb

					Brent Baccala
					cosine@freesoft.org
-
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/