Re: 2.6.1-rc1 [resend]

From: Paolo Ornati (ornati_at_lycos.it)
Date: 12/31/03

  • Next message: Mike Fedyk: "Re: [2.4] Server stops serving nfs files (possibly ext3 or quota related)" 
    To: Ed Sweetman <ed.sweetman@wmich.edu>
Date:	Wed, 31 Dec 2003 19:03:56 +0100

    On Wednesday 31 December 2003 17:31, Ed Sweetman wrote:
    > Paolo Ornati wrote:
    > > On Wednesday 31 December 2003 16:20, you wrote:
    > >>On Wednesday 31 December 2003 16:06, you wrote:
    > >>>>What io scheduler are you using? Or, could you post /var/log/dmesg?
    > >>
    > >>On Wed, Dec 31, 2003 at 04:19:27PM +0100, Paolo Ornati wrote:
    > >>>"dmesg" and "config" attached.
    > >>
    > >>Could you try this with elevator=deadline?
    > >
    > > ok, I have just tried...
    > > I don't see any big difference.
    >
    > Wasn't it mentioned in another thread related to a drop in ide
    > performance that there is possibly some bug in the ide code that ends up
    > requiring you to set the readahead on all your devices to see the max
    > performance of any one?
    >
    > set all the readaheads of all your ide devices to 8192 You should see
    > the best peformance doing this.

    No, I have just try it but I don't see any changes.

    My question is about a strange change of "behaviour" in IDE performance
    changing readahead.

    Here I report a new thing that I've noticed today:

    Kernel 2.6.0:

    o readahead up to 224:
    when I run "hdparm -t /dev/hda" the HD LED light up... and after a while
    light down

    o readahead > 224:
    running "hdparm" the HD LED starts blinking... and then light down

    The best performance are riched with values between 128 and 224 --> IOW when
    the HD led starts blinking the performance diminish...

    Kernel 2.6.1-rc1:

    the HD LED starts to blink with readahead = 16 !!!
    Why?

    BYE 

    -- 
	Paolo Ornati
	Linux v2.6.0
-
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/
  • Next message: Mike Fedyk: "Re: [2.4] Server stops serving nfs files (possibly ext3 or quota related)"

    Relevant Pages

    • Re: Driver retries disk errors.
      ... > IDE layer's job is to make sure it doesn't get lost in the first place. ... > block layer and IDE can abort the readahead sequence happily enough ... to avoid passing back errors caused by readahead. ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)
    • Re: 2.6.1-rc1 [resend]
      ... Wasn't it mentioned in another thread related to a drop in ide ... requiring you to set the readahead on all your devices to see the max ... set all the readaheads of all your ide devices to 8192 You should see ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)
    • Re: very low performance on SCSI disks if device node is in tmpfs
      ... It affects IDE too - the context switch and interrupt rates are through the ... IDE disks have always seemed better at doing readahead, ... Everything there is consistent with "not doing readahead". ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)
    • Re: Strange IDE performance change in 2.6.1-rc1 (again)
      ... beacuse readahead setted to 8192 gives me BAD performance! ... > regression that you are, my dd doesn't do what your dd does. ... > due to a regression in the schedular and nothing to do with the ide ... > changes with your readahead values) then the same kernel with the new ...
      (Linux-Kernel)
    • Re: Strange IDE performance change in 2.6.1-rc1 (again)
      ... I do not see this behavior and i'm using the same ide chipset driver ... readahead for all my other ... drives is set to 8192 during these tests but changing them showed no ... hdparm -t is difficult to get really accurate, ...
      (Linux-Kernel)