VM changes added to performance patches for 2.4.19

VM changes added to performance patches for 2.4.19

[conman]

With the patch against 2.4.19:

Scheduler O(1), Preemptible, Low Latency

I have now added two extra alternative patches that include 
either Rik's rmap (thanks Rik) or AA's vm changes (thanks to Nuno Monteiro for
merging this)

For the record, with the (very) brief usage of these two patches I found the
rmap patch a little faster. This is very subjective and completely untested.

Check them out here and tell me what you think(please read the FAQ):
http://kernel.kolivas.net

Con Kolivas

Re: VM changes added to performance patches for 2.4.19

[Dag Nygren]

> 
> With the patch against 2.4.19:
> 
> Scheduler O(1), Preemptible, Low Latency
> 
> I have now added two extra alternative patches that include 
> either Rik's rmap (thanks Rik) or AA's vm changes (thanks to Nuno Monteiro for
> merging this)
> 
> For the record, with the (very) brief usage of these two patches I found the
> rmap patch a little faster. This is very subjective and completely untested.
> 
> Check them out here and tell me what you think(please read the FAQ):
> http://kernel.kolivas.net

Tried this patch and it applied fine.
The target is a laptop though and I find that this patch will somehow generate
lots of diskaccesses and thus the disk will never "sleep".
Checking out what the reason might be it seems like it is the preempt stuff 
that
writes notes about processes been kicked ouf to /var/log/messages.

Is there a way to shut this off ?

BRGDS and  thanks for the unified patch


-- 
Dag Nygren                               email: dag@newtech.fi
Oy Espoon NewTech Ab                     phone: +358 9 8024910
Träsktorpet 3                              fax: +358 9 8024916
02360 ESBO                              Mobile: +358 400 426312
FINLAND

Re: VM changes added to performance patches for 2.4.19

[Daniel Phillips]

On Saturday 24 August 2002 08:33, conman@kolivas.net wrote:
> With the patch against 2.4.19:
> 
> Scheduler O(1), Preemptible, Low Latency
> 
> I have now added two extra alternative patches that include 
> either Rik's rmap (thanks Rik) or AA's vm changes (thanks to Nuno Monteiro for
> merging this)
> 
> For the record, with the (very) brief usage of these two patches I found the
> rmap patch a little faster. This is very subjective and completely untested.

Would you be so kind as to attempt to quantify that?

-- 
Daniel

Re: VM changes added to performance patches for 2.4.19

[conman]

Quoting Daniel Phillips <phillips@arcor.de>:
> Would you be so kind as to attempt to quantify that?

Ummm... I'm not sure if you're making fun or me? I haven't done any objective
tests so I can't quantify it ??

I just found the responsiveness of the machine a little better and don't have
the resources, time or inclination to test it with a benchmark. It's my
understanding that the -aa patch performed better on benchmarks, but that some
people reported the responsiveness was better with -rmap anyway. I'd agree with
the latter statement. I offer both patches with mine so if people want to try my
patch and feel strongly either way they can choose. My aim is to optimise system
response for single cpu desktops, not multi cpu servers.

Con.

Re: VM changes added to performance patches for 2.4.19

[Daniel Phillips]

On Saturday 24 August 2002 16:36, conman@kolivas.net wrote:
> Quoting Daniel Phillips <phillips@arcor.de>:
> > Would you be so kind as to attempt to quantify that?
> 
> Ummm... I'm not sure if you're making fun or me? I haven't done any objective
> tests so I can't quantify it ??

Not at all.  We need to quantify such results as yours, if we can.

> I just found the responsiveness of the machine a little better and don't have
> the resources, time or inclination to test it with a benchmark.

Fair enough.

> It's my
> understanding that the -aa patch performed better on benchmarks, but that some
> people reported the responsiveness was better with -rmap anyway. I'd agree with
> the latter statement. I offer both patches with mine so if people want to try my
> patch and feel strongly either way they can choose. My aim is to optimise system
> response for single cpu desktops, not multi cpu servers.

-- 
Daniel

Re: VM changes added to performance patches for 2.4.19

[Cliff White]

> Quoting Daniel Phillips <phillips@arcor.de>:
> > Would you be so kind as to attempt to quantify that?
> 
> Ummm... I'm not sure if you're making fun or me? I haven't done any objective
> tests so I can't quantify it ??
> 
> I just found the responsiveness of the machine a little better and don't have
> the resources, time or inclination to test it with a benchmark. It's my
> understanding that the -aa patch performed better on benchmarks, but that some
> people reported the responsiveness was better with -rmap anyway. I'd agree with
> the latter statement. I offer both patches with mine so if people want to try my
> patch and feel strongly either way they can choose. My aim is to optimise system
> response for single cpu desktops, not multi cpu servers.
> 
> Con.
> -
> 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/
> 
Con, 
Thanks for the work, and nice web page!
To help with the testing, i have taken the liberty of adding your 2.4.19-ck3 
patch to the OSDL's Patch
LifeCycle Manager. It's patch # 768. I've queued up a couple of the basic 
tests against it, (dbench, dbt1 )
If you want to add future versions and get some testing run, see 
http://osdl.org/cgi-bin/plm - it's really easy,
and you can use OSDL to make up for your lack of hardware.  We do focus more 
on multi-cpu systems, but
we're always interested in scheduler tests.

cliffw

Re: VM changes added to performance patches for 2.4.19

[Bill Davidsen]

On Sat, 24 Aug 2002 conman@kolivas.net wrote:

> 
> With the patch against 2.4.19:
> 
> Scheduler O(1), Preemptible, Low Latency
> 
> I have now added two extra alternative patches that include 
> either Rik's rmap (thanks Rik) or AA's vm changes (thanks to Nuno Monteiro for
> merging this)
> 
> For the record, with the (very) brief usage of these two patches I found the
> rmap patch a little faster. This is very subjective and completely untested.
> 
> Check them out here and tell me what you think(please read the FAQ):
> http://kernel.kolivas.net

I tried the 2.4.19-ck3-aa patch last night, and did a few informal tests
against my current production kernel, 2.4.19-ac4. Machine in Athlon
1400MHz, 1GB RAM, 20+30GB WD disks.

Kernel compile was about 7 sec faster with ck3-aa, 6:58 vs 7:05 (no -j
values).

Then I did my nightly backup of a scanned documentation project, making a
CD image from the scans, currently ~570MB. I was on ck3-aa, and I said
"self, that seemed pretty fast!" So I rebooted cold and tried the mkisofs
with both kernels, twice each.

			2.4.19-ac4		2.4.19-ck3-aa
mkisofs 570MB		2:05			1:14

It was repeatably 40% faster! More testing, and now I'll build a stock
2.4.19 kernel for additional testing, and pull that responsiveness
benchmark and try that, too.

Looks like a nice job overall, I'm putting it on a laptop tonight, which
may give a better idea of how fast, or not, it really is.

-- 
bill davidsen <davidsen@tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.

Re: VM changes added to performance patches for 2.4.19

[Bill Davidsen]

On Sat, 24 Aug 2002 conman@kolivas.net wrote:

> 
> With the patch against 2.4.19:
> 
> Scheduler O(1), Preemptible, Low Latency
> 
> I have now added two extra alternative patches that include 
> either Rik's rmap (thanks Rik) or AA's vm changes (thanks to Nuno Monteiro for
> merging this)
> 
> For the record, with the (very) brief usage of these two patches I found the
> rmap patch a little faster. This is very subjective and completely untested.
> 
> Check them out here and tell me what you think(please read the FAQ):
> http://kernel.kolivas.net

The ck3-aa patch has worked perfectly for me until I try to shut down. At
that point I get to "turning off swap" and the system hangs with the disk
light on. Can't get a dump, and it doesn't happen every time, but enough
that I am very cautious in what I do at shutdown. Total hang ignoring
sysreq.

Athlon 1.4GHz, 1GB RAM, hda:30GB, hdc:40GB, 20x CD-R, multiple NICs, two
local networks, one PPP over high speed serial.

-- 
bill davidsen <davidsen@tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.

Re: VM changes added to performance patches for 2.4.19

[Con Kolivas]

Quoting Bill Davidsen <davidsen@tmr.com>:
> On Sat, 24 Aug 2002 conman@kolivas.net wrote:
> > With the patch against 2.4.19:
> > Scheduler O(1), Preemptible, Low Latency
> > I have now added two extra alternative patches that include 
> > either Rik's rmap (thanks Rik) or AA's vm changes (thanks to Nuno Monteiro
> for
> > merging this)
> > For the record, with the (very) brief usage of these two patches I found
> the
> > rmap patch a little faster. This is very subjective and completely
> untested.
> > Check them out here and tell me what you think(please read the FAQ):
> > http://kernel.kolivas.net
> 
> The ck3-aa patch has worked perfectly for me until I try to shut down. At
> that point I get to "turning off swap" and the system hangs with the disk
> light on. Can't get a dump, and it doesn't happen every time, but enough
> that I am very cautious in what I do at shutdown. Total hang ignoring
> sysreq.
> 
> Athlon 1.4GHz, 1GB RAM, hda:30GB, hdc:40GB, 20x CD-R, multiple NICs, two
> local networks, one PPP over high speed serial.

Check on the website and you'll see that there have been two upgrades. The -ck5
patch now includes the -aa vm changes by default, and the hang on shutdown (due
to swapoff failing) has been fixed. For the record, the ck5 patch is the last
one until a new compressed cache patch becomes available for 2.4.19 and I will
merge that to make a -ck6 patch. The ck5 patch has proven to be very stable and
I am satisfied that it needs no further changes till the compressed cache patch
becomes available, so I recommend you upgrade to that.

Regards,
Con Kolivas.

Re: VM changes added to performance patches for 2.4.19

[Marc-Christian Petersen]

Hi Bill,

> ...
> against my current production kernel, 2.4.19-ac4. Machine in Athlon
> 1400MHz, 1GB RAM, 20+30GB WD disks.
> ...
> Then I did my nightly backup of a scanned documentation project, making a
> CD image from the scans, currently ~570MB. I was on ck3-aa, and I said
> "self, that seemed pretty fast!" So I rebooted cold and tried the mkisofs
> with both kernels, twice each.

>                         2.4.19-ac4              2.4.19-ck3-aa
> mkisofs 570MB           2:05                    1:14


with WOLK v3.6 and only ONE harddisk drive:

root@codeman:[/] # du -skh /home
605M    /home
root@codeman:[/] # find /home|wc -l
   7384
root@codeman:[/] # time mkisofs -o /delme.iso -allow-multidot -iso-level 3 -J 
-max-iso9660-filenames -U -R /home
...

real    1m12.872s
user    0m1.840s
sys     0m10.800s


with one big file:

root@codeman:[/] # du -skh /home/onebigfile.delme
620M    /home
root@codeman:[/] # time mkisofs -o /delme.iso -allow-multidot -iso-level 3 -J 
-max-iso9660-filenames -U -R /home/onebigfile.delme
...

real    0m35.676s
user    0m1.180s
sys     0m10.740s

:-)

Anyway, I also tested the patchset from Con, both, with -aa VM and -rmap VM 
and I must say, -rmap is much slower than -aa VM.


Testmashine: Celeron 800MHz
             256MB RAM
             1x Seagate 30GB running in UDMA4 (ATA66) mode
             Filesystem: ext3, data=ordered

I think with your mashine this could be a bit faster also ;)

-- 
Kind regards
        Marc-Christian Petersen

http://sourceforge.net/projects/wolk

PGP/GnuPG Key: 1024D/569DE2E3DB441A16
Fingerprint: 3469 0CF8 CA7E 0042 7824 080A 569D E2E3 DB44 1A16
Key available at www.keyserver.net. Encrypted e-mail preferred.