Re: [RFC] [PATCH] 2.5.35 patch for making DIO async

Badari Pulavarty (pbadari@us.ibm.com)
Wed, 18 Sep 2002 08:58:22 -0700 (PDT)


>
> On Tue, Sep 17, 2002 at 02:03:02PM -0700, Badari Pulavarty wrote:
> > Hi Ben,
> >
> > Here is a 2.5.35 patch to make DIO async. Basically, DIO does not
> > wait for io completion. Waiting will be done at the higher level
> > (same way generic_file_read does).
>
> This looks pretty good. Andrew Morton has had a look over it too and
> agrees that it should go in after a bit of testing. Does someone want
> to try comparing throughput of dio based aio vs normal read/write? It
> would be interesting to see what kind of an effect pipelining aios makes
> and if there are any performance niggles we need to squash. Cheer,
>
> -ben

Thanks for looking at the patch. Patch needed few cleanups to get it
working. Here is the status so far..

1) I tested synchronous raw read/write. They perform almost same as
without this patch. I am looking at why I can't match the performance.
(I get 380MB/sec on 40 dd's on 40 disks without this patch.
I get 350MB/sec with this patch).

2) wait_on_sync_kiocb() needed blk_run_queues() to make regular read/
write perform well.

3) I am testing aio read/writes. I am using libaio.0.3.92.
When I try aio_read/aio_write on raw device, I am get OOPS.
Can I use libaio.0.3.92 on 2.5 ? Are there any interface
changes I need to worry between 2.4 and 2.5 ?

Once I get aio read/writes working, I will provide you the performance
numbers.

Thanks,
Badari

-
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/