test8-vmpatch performs great here!

test8-vmpatch performs great here!

[deprogrammer]

After reading ben's email I dicided to run his same test on my box
running test8 + vmpatch3 

some specs: K7 - 600, 128MB ram.

environment: X 4.0.1 and netscape 4.75 were running.

I ran 2 tests:
	1) tar zxvf linux-2.4.0-test6.tar.gz
	2) tar xvf linux-2.4.0-test6.tar


free_before:
	total       used       free     shared    buffers     cached
	Mem:        127176      62960      64216          0       2164      25500
	-/+ buffers/cache:      35296      91880
	Swap:       128516          0     128516

free_after_tgz:
	total       used       free     shared    buffers     cached
	Mem:        127176     124892       2284          0       6612      80592
	-/+ buffers/cache:      37688      89488
	Swap:       128516          0     128516

free_after_tar:
total       used       free     shared    buffers     cached
Mem:        127176     124952       2224          0       2848      85336
-/+ buffers/cache:      36768      90408
Swap:       128516          0     128516

The box remained somewhat interactive, but a few times during the tar zxvf the
box would stop responding for a few seconds during which there would be alot
of disk activity, same for the tar xvf.

john
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: test8-vmpatch performs great here!

[Rik van Riel]

On Fri, 8 Sep 2000, deprogrammer wrote:

> The box remained somewhat interactive, but a few times during
> the tar zxvf the box would stop responding for a few seconds
> during which there would be alot of disk activity, same for the
> tar xvf.

This seems to be due to kflushd (bdflush) waking up
tasks in LIFO order ...

>From fs/buffer.c:

in wakeup_bdflush()
   2435         __set_current_state(TASK_UNINTERRUPTIBLE);
   2436         add_wait_queue(&bdflush_done, &wait);
   2437 
   2438         wake_up_process(bdflush_tsk);
   2439         schedule();

(which adds our task to the front of the wait queue)

and in kflushd()
   2622                 wake_up(&bdflush_done);

(which wakes up the first task on the wait queue)

This results in LIFO ordering for the wakeup. I've
fixed this in my local tree (doing a wake_up_all()
instead) and will release a new patch after tweaking
things a bit more.

regards,

Rik
--
"What you're running that piece of shit Gnome?!?!"
       -- Miguel de Icaza, UKUUG 2000

http://www.conectiva.com/		http://www.surriel.com/

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: test8-vmpatch performs great here!

[Martin Josefsson]

On Fri, 8 Sep 2000, deprogrammer wrote:

> 
> After reading ben's email I dicided to run his same test on my box
> running test8 + vmpatch3 
> 
> some specs: K7 - 600, 128MB ram.
> 
> environment: X 4.0.1 and netscape 4.75 were running.
> 
> I ran 2 tests:
> 	1) tar zxvf linux-2.4.0-test6.tar.gz
> 	2) tar xvf linux-2.4.0-test6.tar
> 
> 
> free_before:
> 	total       used       free     shared    buffers     cached
> 	Mem:        127176      62960      64216          0       2164      25500
> 	-/+ buffers/cache:      35296      91880
> 	Swap:       128516          0     128516
> 
> free_after_tgz:
> 	total       used       free     shared    buffers     cached
> 	Mem:        127176     124892       2284          0       6612      80592
> 	-/+ buffers/cache:      37688      89488
> 	Swap:       128516          0     128516
> 
> free_after_tar:
> total       used       free     shared    buffers     cached
> Mem:        127176     124952       2224          0       2848      85336
> -/+ buffers/cache:      36768      90408
> Swap:       128516          0     128516
> 
> The box remained somewhat interactive, but a few times during the tar zxvf the
> box would stop responding for a few seconds during which there would be alot
> of disk activity, same for the tar xvf.

I have to say that the vm-patch against t8p5 works fine here on t8p6

I've tested to copy a few big files between partitions and it didn't swap
at all (it used to swap like crazy leaving me to watch my frozen X during
the copy)

so far so good....

then a friend started a ftp-session against my machine and started to copy
a few files (3-6MB each via 100Mbit network) and my machine began swapping
like there was no tomorrow.

after about 15-30 seconds of heavy swapping the swapping stopped, now I
had about 35MB swapped out (I have 256MB ram, and the box is very light
loaded, no big memoryhogs or anything (not even netscape :))

I took a look at /proc/meminfo and I had more than 200MB in active pages
and almost nothing in inactive pages. So this is a case where drop-behind
doesn't work yet.

I should say that my ftpserver is proftpd-1.2.0pre10, a quite common
ftpserver

/Martin

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: test8-vmpatch performs great here!

[Stephen C. Tweedie]

Hi,

On Fri, Sep 08, 2000 at 08:29:43PM -0300, Rik van Riel wrote:

> This seems to be due to kflushd (bdflush) waking up
> tasks in LIFO order ...

Hmm??

> >From fs/buffer.c:
> 
> in wakeup_bdflush()
>    2435         __set_current_state(TASK_UNINTERRUPTIBLE);
>    2436         add_wait_queue(&bdflush_done, &wait);

> (which adds our task to the front of the wait queue)

Right, but it is TASK_UNINTERRUPTIBLE, not TASK_EXCLUSIVE.

> and in kflushd()
>    2622                 wake_up(&bdflush_done);
> 
> (which wakes up the first task on the wait queue)

No.  That might be true if we were TASK_EXCLUSIVE, but we are not ---
*all_ processes on the wait queue will be woken, and the scheduler
doesn't care in the slightest about which order they get woken in.  It
will just schedule the best process next time.
 
wake_up_all() is only different from wake_up() when you encounter
TASK_EXCLUSIVE processes.

Cheers,
 Stephen
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: test8-vmpatch performs great here!

[Rik van Riel]

On Sat, 9 Sep 2000, Stephen C. Tweedie wrote:
> On Fri, Sep 08, 2000 at 08:29:43PM -0300, Rik van Riel wrote:
> 
> > >From fs/buffer.c:
> 
> > and in kflushd()
> >    2622                 wake_up(&bdflush_done);
> > 
> > (which wakes up the first task on the wait queue)
> 
> No.  That might be true if we were TASK_EXCLUSIVE, but we are
> not --- *all_ processes on the wait queue will be woken,

> wake_up_all() is only different from wake_up() when you
> encounter TASK_EXCLUSIVE processes.

Indeed, this fooled me. For readability I have changed
the wake_up() to a wake_up_all() anyway ;)

regards,

Rik
--
"What you're running that piece of shit Gnome?!?!"
       -- Miguel de Icaza, UKUUG 2000

http://www.conectiva.com/		http://www.surriel.com/

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/