Re: [PATCH] rmap 15j for 2.4.21-rc6

Re: [PATCH] rmap 15j for 2.4.21-rc6

[ghugh Song]

Michael Frank wrote:

>On a P4/533-2.4Ghz/512MB with udma5 IDE ~50MB/s:
>
>Shows severe interactivity problems and hangs
>
>Scroll and mouse hangs and delayed response to keyboard
>greater 1s are easily observable.
>
>Test script: tstinter V0.1

Unlike your case,
On a Athlon 2700+/166MHz DDR/1GB on a board having Nvidia2 chipset
with SCSI hard disks, it solved all my interactivity problem.
BTW, I do LaTeXing the most.  I have never tested such test
scripts.

More information about my case can be found in

http://www.ussg.iu.edu/hypermail/linux/kernel/0305.3/1894.html

Regards,

G. Hugh Song

Re: [PATCH] rmap 15j for 2.4.21-rc6

[Michael Frank]

On Sunday 01 June 2003 12:57, ghugh Song wrote:

> On a Athlon 2700+/166MHz DDR/1GB on a board having Nvidia2 chipset
> with SCSI hard disks, it solved all my interactivity problem.
> BTW, I do LaTeXing the most.

Congratulations on winning at the roulette table of everyday use ;-)

> I have never tested such test  scripts.

How many hours of users and developers time have been wasted on 
"interactivity problems, locks and hangs" since 2.4.18 ?

Regards
Michael 

Re: [PATCH] rmap 15j for 2.4.21-rc6

[Andreas Hartmann]

Hi,

Willy Tarreau wrote:

> Hi !
> 
> On Mon, Jun 02, 2003 at 06:55:48PM +0200, Andreas Hartmann wrote:
>> Michael Frank wrote:
> <...>
>> > GNU bash, version 2.05b.0(1)-release (i386-redhat-linux-gnu)
>> 
>> 2.02.1(1)-release
> <...>
>> > -   while (( i-- )); do
>> > +   while (( i=`expr $i - 1` )); do
>> > 
>> > In your opinion are your changes more portable across a wide range of
>> > systems?
>> 
>> I didn't think at portability :-). I only made it working for me. Maybe
>> there are other persons out there who do have some old versions too - so
>> they can use this patch.
> 
> Well, I found that i--/i++ don't work with bash-2.03 (present about
> everywhere) but i=i-1 or i=i+1 work well. So at least, for portability,
> this could be rewritten as "while (( i=i-1 )); do".

Works fine with 2.02.1, too.


Regards,
Andreas Hartmann

Re: [PATCH] rmap 15j for 2.4.21-rc6

[Willy Tarreau]

Hi !

On Mon, Jun 02, 2003 at 06:55:48PM +0200, Andreas Hartmann wrote:
> Michael Frank wrote:
<...>
> > GNU bash, version 2.05b.0(1)-release (i386-redhat-linux-gnu)
> 
> 2.02.1(1)-release
<...>
> > -   while (( i-- )); do
> > +   while (( i=`expr $i - 1` )); do
> > 
> > In your opinion are your changes more portable across a wide range of
> > systems?
> 
> I didn't think at portability :-). I only made it working for me. Maybe
> there are other persons out there who do have some old versions too - so
> they can use this patch.

Well, I found that i--/i++ don't work with bash-2.03 (present about everywhere)
but i=i-1 or i=i+1 work well. So at least, for portability, this could be
rewritten as "while (( i=i-1 )); do".

Cheers,
Willy

Re: [PATCH] rmap 15j for 2.4.21-rc6

[Andreas Hartmann]

Michael Frank wrote:

> On Sunday 01 June 2003 19:00, Andreas Hartmann wrote:
>>
>> Well, I did the test with 2.4.21rc6 after patching your script (I got
>> syntax errors):
> 
> About your script changes, I like to make it portable, and I use the
> following versions:
> 
> GNU bash, version 2.05b.0(1)-release (i386-redhat-linux-gnu)

2.02.1(1)-release

> 
> dd (coreutils) 4.5.3

dd (fileutils) 4.1

> What shell and coreutils are you using?

They are obviously much older :-).

> Avoiding short counts is easy but avoiding C-style expressions is
> primitive
> 
> -     count=100K
> +     count=100000
> 
> -   while (( i-- )); do
> +   while (( i=`expr $i - 1` )); do
> 
> In your opinion are your changes more portable across a wide range of
> systems?

I didn't think at portability :-). I only made it working for me. Maybe
there are other persons out there who do have some old versions too - so
they can use this patch.

>> When I'm using the script as seen in the patch, I'm getting problems with
>> df (it's mostly very lazy, about 20s delay or more), the load is 4, doing
>> an ls on some other directories is extremly slow. Mouse and keyboard are
>> hanging some times.
>> The write speed shown in xosview was between 1 and 15MB/s. Often the HD
>> LED was on, but no data seemed to be put to the HD.
>>
> 
> It has a hard time to read anything else, the slower the disk, the worse.
> 
> Suppose rmap undoes the fixes introduced in -rc6.
> 
> Have you tried -rc6 plain?

Yes - I only tested this kernel.



Regards,
Andreas Hartmann

Re: [PATCH] rmap 15j for 2.4.21-rc6

[Rik van Riel]

On Sun, 1 Jun 2003, Michael Frank wrote:

> Suppose rmap undoes the fixes introduced in -rc6. 

It shouldn't, unless bitkeeper made a serious merge
error (unlikely).  The rmap patch should just change
VM code, not the IO improvements.

Having said that, I'd like to thank Michael Frank for
his script that reproduces problems.  Now the problem
is reproducable I can try to fix it ...

Re: [PATCH] rmap 15j for 2.4.21-rc6

[Michael Frank]

On Sunday 01 June 2003 19:00, Andreas Hartmann wrote:
>
> Well, I did the test with 2.4.21rc6 after patching your script (I got
> syntax errors):

About your script changes, I like to make it portable, and I use the 
following versions:

GNU bash, version 2.05b.0(1)-release (i386-redhat-linux-gnu)

dd (coreutils) 4.5.3

What shell and coreutils are you using?

Avoiding short counts is easy but avoiding C-style expressions is primitive

-     count=100K
+     count=100000

-   while (( i-- )); do
+   while (( i=`expr $i - 1` )); do

In your opinion are your changes more portable across a wide range of systems? 

>
> When I'm using the script as seen in the patch, I'm getting problems with
> df (it's mostly very lazy, about 20s delay or more), the load is 4, doing
> an ls on some other directories is extremly slow. Mouse and keyboard are
> hanging some times.
> The write speed shown in xosview was between 1 and 15MB/s. Often the HD LED
> was on, but no data seemed to be put to the HD.
>

It has a hard time to read anything else, the slower the disk, the worse. 

Suppose rmap undoes the fixes introduced in -rc6. 

Have you tried -rc6 plain?


Regards
Michael

Re: [PATCH] rmap 15j for 2.4.21-rc6

[Andreas Hartmann]

Michael Frank wrote:

> On Saturday 31 May 2003 22:48, Rik van Riel wrote:
>>
>> Today I finally merged rmap15j forward to marcelo's latest
>> release.  The IO stall fixes should be especially interesting:
>>
> 
> Patched rc6 ex BK OK and compiled with gcc295-3 OK
> 
> On a P4/533-2.4Ghz/512MB with udma5 IDE ~50MB/s:
> 
> Shows severe interactivity problems and hangs
> 
> Scroll and mouse hangs and delayed response to keyboard
> greater 1s are easily observable.
> 

Well, I did the test with 2.4.21rc6 after patching your script (I got syntax
errors):

When I'm using the script as seen in the patch, I'm getting problems with df
(it's mostly very lazy, about 20s delay or more), the load is 4, doing an
ls on some other directories is extremly slow. Mouse and keyboard are
hanging some times.
The write speed shown in xosview was between 1 and 15MB/s. Often the HD LED
was on, but no data seemed to be put to the HD.


My system:
AMD Athlon(tm) XP 2000+
512 MB RAM
Chipsatz Apollo KT266/A/333
WDC WD205AA, udma4, 20,5 GB
             (multcount    = 16 (on),I/O support  =  1 (32-bit),
             unmaskirq    =  1 (on),using_dma    =  1 (on),
             readahead    =  8 (on)
             )
reiserfs


Regards,
Andreas Hartmann


--- /home/Andreas/tstinter      Sun Jun  1 11:03:10 2003
+++ tstinter    Sun Jun  1 11:27:18 2003
@@ -138,7 +138,7 @@
     sleep 1
    fi
  #invoke dd write loops
-   while (( i-- )); do
+   while (( i=`expr $i - 1` )); do
      $TSTINTER _xwrite "$2" $i
      sleep 1
    done
@@ -163,13 +163,13 @@
  #
  _read)
    if [ "$2" = "" ]; then
-     count=100K
+     count=100000
    else
      count=$2
    fi
-   dd if=/dev/zero of=$TEMPFILE$3 bs=4K count=$count
+   dd if=/dev/zero of=$TEMPFILE$3 bs=4096 count=$count
    while (( 1 )); do
-     time dd if=$TEMPFILE$3 of=/dev/null bs=4K count=$count &> /dev/null
+     time dd if=$TEMPFILE$3 of=/dev/null bs=4096 count=$count &> /dev/null
    done
    ;;
  #
@@ -181,12 +181,12 @@
  #
  _write)
    if [ "$2" = "" ]; then
-     count=100K
+     count=100000
    else
      count=$2
    fi
    while (( 1 )); do
-     time dd if=/dev/zero of=$TEMPFILE$3 bs=4K count=$count
+     time dd if=/dev/zero of=$TEMPFILE$3 bs=4096 count=$count
    done
    ;;
  #

Re: [PATCH] rmap 15j for 2.4.21-rc6

[Michael Frank]

On Saturday 31 May 2003 22:48, Rik van Riel wrote:
>
> Today I finally merged rmap15j forward to marcelo's latest
> release.  The IO stall fixes should be especially interesting:
>

Patched rc6 ex BK OK and compiled with gcc295-3 OK

On a P4/533-2.4Ghz/512MB with udma5 IDE ~50MB/s:

Shows severe interactivity problems and hangs

Scroll and mouse hangs and delayed response to keyboard 
greater 1s are easily observable.

Test script: tstinter V0.1 

http://www.ussg.iu.edu/hypermail/linux/kernel/0305.3/1291.html

To reproduce cd to dir with script and execute from X console: 
  ./tstinter start

More instructions in script

Other kernel results: 
   2.4.18 PIO DIES - see msg w. script, 
   2.4.18 udma2 OK, 
   2.4.19 Bad, 
   2.4.20 Bad, 
   2.4.21-rc1 Bad, 
   2.4.21-rc6 OK 
   2.5.70,-mm1,-mm2,-mm3 (OK) 

Regards
Michael

[PATCH] rmap 15j for 2.4.21-rc6

[Rik van Riel]

On Fri, 30 May 2003, Rik van Riel wrote:

> The tenth maintenance release of the 15th version of the reverse
> mapping based VM is now available.
> This is an attempt at making a more robust and flexible VM
> subsystem, while cleaning up a lot of code at the same time.
> The patch is available from:
> 
>            http://surriel.com/patches/2.4/2.4.21-pre7-rmap15j
> and        http://linuxvm.bkbits.net/

Today I finally merged rmap15j forward to marcelo's latest
release.  The IO stall fixes should be especially interesting:

http://surriel.com/patches/2.4/2.4.21-rc6-rmap15j