RE: 2.4.26 and MPC885 context switching time

RE: 2.4.26 and MPC885 context switching time

[Schaefer-Hutter, Peter]

> Maybe it would help if you gave more detail about how
> you are measuring these times in your system.

I basically used the two programs from=20
http://www.ibm.com/developerworks/linux/library/l-rt9/
since running lmbench is currently not an option;
however, i'm looking into creating a nfs filesystem that
has enough stuff in it to run lmbench.

  Regards,

    Peter

Re: 2.4.26 and MPC885 context switching time

[Dan Malek]

On Jun 15, 2005, at 11:43 AM, Schaefer-Hutter, Peter wrote:

> I basically used the two programs from
> http://www.ibm.com/developerworks/linux/library/l-rt9/

That's a fine test, if you run it on various systems and
compare the _relative_ results.  There is more than
context switching going on here.

.... there aren't any MMU "problems" with the 8xx, it's
likely to be the most efficient implementation of all
systems that use software to support the MMU.  The
processor in general is very sensitive to many system
architecture decisions, especially the system bus
and DRAM configuration.  Make certain you really
understand what you have configured and how the
tests you run are affected by that.

Thanks.


	-- Dan

RE: 2.4.26 and MPC885 context switching time

[Steven Blakeslee]

> Therefore i wonder, if there are any known problems with the=20
> MMU on 8xx and 2.4.2x which might explain that?=20
>=20

I did some work on an 885 with Linux 2.4.22.  I found that when the
processor was running at 100MHz or higher Linux would die when
decompressing the ramdisk.  Below 100MHz it would run fine.  I came up
with a hacky fix because I mis-read an errata notice.  What I did was
set the ICTRL register to 0x00000001, this fixed the problem.  Basically
all this does is slow the processor down considerable.  The Linux you
have may be doing the same thing.  This could explain the performance
hit you are seeing.

RE: 2.4.26 and MPC885 context switching time

[Andrew Williams]

[-- Attachment #1: Type: text/plain, Size: 1235 bytes --]

Question embedded below..

> -----Original Message-----
> From: linuxppc-embedded-bounces@ozlabs.org 
> [mailto:linuxppc-embedded-bounces@ozlabs.org] On Behalf Of 
> Martin Krause
> Sent: June 15, 2005 11:12 AM
> To: Schaefer-Hutter, Peter
> Cc: linuxppc-embedded@ozlabs.org
> Subject: AW: 2.4.26 and MPC885 context switching time
> 
> 
> Hi Peter,
> 
> linuxppc-embedded-bounces@ozlabs.org wrote on :
> > i'm evaluating IPC on the MPC885ADS and discovered
> > that the context switching times seem to be quite 
> disappointing (up to 
> > 1ms). As we need to do a lot
> .. 
> > Or is this the expected context switching time for
> > CPU @ 100 MHz (RAM @ 50 MHz)? Suggestions for a
> > benchmark program?
> 
> Denx has done some detailed performance measurements 
> (including context switching). You could find the report 
> here: http://www.denx.de/twiki/bin/view/Know/Linux24vs26.
> Hope that is of some help.
>

Has anything changed with 2.6 in the last 12 months that 
would alter the conclusions presented in this article?

A.

 
> Regards,
> Martin
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org 
> https://ozlabs.org/mailman/listinfo/linuxppc-> embedded
> 
> 

[-- Attachment #2: Type: text/html, Size: 2922 bytes --]

2.4.26 and MPC885 context switching time

[Schaefer-Hutter, Peter]

Hi all,

i'm evaluating IPC on the MPC885ADS and discovered
that the context switching times seem to be quite=20
disappointing (up to 1ms). As we need to do a lot=20
of IPC later on, i'm interested bringing the time=20
down.

Therefore i wonder, if there are any known problems=20
with the MMU on 8xx and 2.4.2x which might explain=20
that?=20

Or is this the expected context switching time for=20
CPU @ 100 MHz (RAM @ 50 MHz)? Suggestions for a=20
benchmark program?

Thanks!

  Peter

Re: 2.4.26 and MPC885 context switching time

[Mark Chambers]

>Hi all,
>
>i'm evaluating IPC on the MPC885ADS and discovered
>that the context switching times seem to be quite 
>disappointing (up to 1ms). As we need to do a lot 
>of IPC later on, i'm interested bringing the time 
>down.
>
>Therefore i wonder, if there are any known problems 
>with the MMU on 8xx and 2.4.2x which might explain 
>that? 

Maybe it would help if you gave more detail about how
you are measuring these times in your system.

I'm interested to know the answer to this myself!

Mark Chambers

Re: 2.4.26 and MPC885 context switching time

[Jaap-Jan Boor]

I did once ran lmbench on a proprietry 850 based board running at  
62/31 Mhz
using 1 16-bit dram:

Context switching - times in microseconds - smaller is better
-------------------------------------------------------------
Host                 OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/ 
64K
                         ctxsw  ctxsw  ctxsw ctxsw  ctxsw   ctxsw    
ctxsw
--------- ------------- ----- ------ ------ ------ ------ -------  
-------
test    Linux 2.4.25- 259.2  234.2         302.6          304.7

This is ~ 3x faster then your 100Mhz 885...

Jaap-Jan

On 15-jun-2005, at 16:55, Schaefer-Hutter, Peter wrote:



> Hi all,
>
> i'm evaluating IPC on the MPC885ADS and discovered
> that the context switching times seem to be quite
> disappointing (up to 1ms). As we need to do a lot
> of IPC later on, i'm interested bringing the time
> down.
>
> Therefore i wonder, if there are any known problems
> with the MMU on 8xx and 2.4.2x which might explain
> that?
>
> Or is this the expected context switching time for
> CPU @ 100 MHz (RAM @ 50 MHz)? Suggestions for a
> benchmark program?
>
> Thanks!
>
>   Peter
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded
>
>
>

____
J.G.J. Boor                       Anton Philipsweg 1
Software Engineer                 1223 KZ Hilversum
AimSys bv                         tel. +31 35 689 1941
Postbus 2194, 1200 CD Hilversum   jjboor at aimsys dot nl

Re: 2.4.26 and MPC885 context switching time

[Wolfgang Denk]

In message <187119B8-515D-4903-AAB5-05BF80215A4D@aimsys.nl> you wrote:
> I did once ran lmbench on a proprietry 850 based board running at  
> 62/31 Mhz using 1 16-bit dram:
> 
> Context switching - times in microseconds - smaller is better
> -------------------------------------------------------------
> Host                 OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/ 
> 64K
>                          ctxsw  ctxsw  ctxsw ctxsw  ctxsw   ctxsw    
> ctxsw
> --------- ------------- ----- ------ ------ ------ ------ -------  
> -------
> test    Linux 2.4.25- 259.2  234.2         302.6          304.7
> 
> This is ~ 3x faster then your 100Mhz 885...

I just ran lmbench on a MPC866 system (CPU at 132 MHz, bus at 66 MHz):

Context switching - times in microseconds - smaller is better
-------------------------------------------------------------------------
Host                 OS  2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
                         ctxsw  ctxsw  ctxsw ctxsw  ctxsw   ctxsw   ctxsw
--------- ------------- ------ ------ ------ ------ ------ ------- -------
tqm8xx     Linux 2.4.25 5.4500   27.5   30.3   39.0   36.0    37.7    33.5


There is definitely something wrong with Peter's measurements.

Best regards,

Wolfgang Denk

-- 
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@denx.de
Quantum particles: The dreams that stuff is made of.