Re: Significant Slowdown Occuring in 2.2 starting with 19pre2

Re: Significant Slowdown Occuring in 2.2 starting with 19pre2

[Andi Kleen]

Alan Cox <alan@lxorguk.ukuu.org.uk> writes:
> 
> The only change in 2.2.19pre2 is the merge of Andrea Arcangeli's VM. Please
> talk to Andrea and see if he can work out why

Also the merge of blkdev-in-pagecache. If the magneto optical disk
has a weird blocksize < PAGE_CACHE_SIZE (doesn't it have 2k normally?) 
then there could be problems.

-Andi

Re: Significant Slowdown Occuring in 2.2 starting with 19pre2

[David S. Miller]

   From: Andi Kleen <ak@suse.de>
   Date: 15 Jan 2002 13:49:16 +0100

   Alan Cox <alan@lxorguk.ukuu.org.uk> writes:
   > 
   > The only change in 2.2.19pre2 is the merge of Andrea Arcangeli's VM. Please
   > talk to Andrea and see if he can work out why
   
   Also the merge of blkdev-in-pagecache.

Andi, 2.2.x not 2.4.x :-)

Re: Significant Slowdown Occuring in 2.2 starting with 19pre2

[Andi Kleen]

On Tue, Jan 15, 2002 at 04:52:44AM -0800, David S. Miller wrote:
>    From: Andi Kleen <ak@suse.de>
>    Date: 15 Jan 2002 13:49:16 +0100
> 
>    Alan Cox <alan@lxorguk.ukuu.org.uk> writes:
>    > 
>    > The only change in 2.2.19pre2 is the merge of Andrea Arcangeli's VM. Please
>    > talk to Andrea and see if he can work out why
>    
>    Also the merge of blkdev-in-pagecache.
> 
> Andi, 2.2.x not 2.4.x :-)

Oops. Thank you David. 

-Andi

Re: Significant Slowdown Occuring in 2.2 starting with 19pre2

[Steve Sheftic]

> 
> Also the merge of blkdev-in-pagecache. If the magneto optical disk
> has a weird blocksize < PAGE_CACHE_SIZE (doesn't it have 2k normally?)
> then there could be problems.
> 

I neglected to mention that. Yes, the media I am using has a 2k
blocksize.

And 2.4.17 is working great - this is only happening with recent 2.2.

Steve

Significant Slowdown Occuring in 2.2 starting with 19pre2

[Steve Sheftic]

Hi,

I encountered a significant disk-write slowdown in the 2.2 kernel that
I've tracked to 2.2.19pre2. I do backups to a SCSI magneto-optical disk.
My /home backup creates a 700MB+ file (using BRU). When I was using
2.2.14, this took roughly a half hour. When I upgraded to 2.2.20, this
same backup started taking nearly 3 hours. Yes, same 700MB+ file size.
The table below shows the results of many tests with many kernels. For
2.2.14 through 2.2.19, the config is essentially identical (doing "make
oldconfig" and answering "No" to new stuff). The config for 2.2.20 and
2.2.21p2 is slightly different, but doesn't seem to be a factor. As you
can see, something happened in 2.2.19pre2 that led to this slowdown.
During these nearly 3 hour backups, the load number stayed at about 3
(with no user-level activity) and the Select light on the MO drive
glowed constantly. System response was poor during this time.

Minutes required to write MO disk vs. kernel version
(multiple tests)

Kernel    Minutes
------    -------
2.2.14     29  33
2.2.15     31  30
2.2.16     67  78       <--- elapsed time more than doubled
2.2.17     22  23  22   <--- notable improvement
2.2.18     34  40
2.2.19p1   30  44
2.2.19p2  161 165       <--- 4x to 5x longer
2.2.19p5  168           (pre3,4 skipped - wouldn't build for me)
2.2.19    168 168
2.2.20    171
2.2.21p2  166

2.4.17     17  17       <--- Woohoo!! 2.4 is GREAT!!

Hardware
--------
Intel 430TX motherboard
Pentium (MMX)
192 MB memory
Advansys ABP940 SCSI controller (hosts the hard disks and MO drive)
Adaptec 2940 SCSI controller (hosts a scanner only)
(more details if requested)

The great news is that for 2.4.17, this backup is only taking 17
minutes! Now that I'm using 2.4, I'm ecstatic! :) However, I wanted to
provide my timing results regarding 2.2.

I also want to express my gratitude to all of you who contribute to this
marvelous OS that I enjoy so much. Thank you!

Steve Sheftic

Re: Significant Slowdown Occuring in 2.2 starting with 19pre2

[Alan Cox]

> I've tracked to 2.2.19pre2. I do backups to a SCSI magneto-optical disk.
> My /home backup creates a 700MB+ file (using BRU). When I was using
> 2.2.14, this took roughly a half hour. When I upgraded to 2.2.20, this

> 2.2.17     22  23  22   <--- notable improvement
> 2.2.18     34  40
> 2.2.19p1   30  44
> 2.2.19p2  161 165       <--- 4x to 5x longer

The only change in 2.2.19pre2 is the merge of Andrea Arcangeli's VM. Please
talk to Andrea and see if he can work out why

Alan