hackbench: 2.6.12-rc6 vs. 2.6.12-rc6-RT-V0.7.48-06

hackbench: 2.6.12-rc6 vs. 2.6.12-rc6-RT-V0.7.48-06

[Gabor MICSKO]

[-- Attachment #1: Type: text/plain, Size: 763 bytes --]

Hi!

I made some test with 2.6.12-rc6-RT-V0.7.48-06 and 2.6.12-rc6 kernels.
Here is my results:

2.6.12-rc6-RT-V0.7.48-0			       2.6.12-rc6
------------------------------------			    ----------------	

./hackbench 30

1st run:  6.139					    5.110	
2nd run: 6.119					   4.946
3rd run: 6.135					    5.168

./hackbench 100

1st run:  23.254				   16.603	
2nd run: 23.481					  16.478
3rd run:  23.790				   16.387

./hackbench 130

1st run:  33.395				   21.731	
2nd run: 32.652					  21.821
3rd run:  32.517				   21.698

./hackbench 150

1st run:  89.100				   47.862	
2nd run: 39.308					  25.121
3rd run:  90.157				   25.125

It seems to me 2.6.12-rc faster than 2.6.12-rc6-RT-V0.7.48-0. It is
normal? 

thanks,

-
mg

[-- Attachment #2: Ez az üzenetrész digitális aláírással van ellátva --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: hackbench: 2.6.12-rc6 vs. 2.6.12-rc6-RT-V0.7.48-06

[Esben Nielsen]

On Fri, 10 Jun 2005, Gabor MICSKO wrote:

> Hi!
> 
> I made some test with 2.6.12-rc6-RT-V0.7.48-06 and 2.6.12-rc6 kernels.
> Here is my results:
> 
> 2.6.12-rc6-RT-V0.7.48-0			       2.6.12-rc6
> ------------------------------------			    ----------------	
> 
> ./hackbench 30
> 
> 1st run:  6.139					    5.110	
> 2nd run: 6.119					   4.946
> 3rd run: 6.135					    5.168
> 
> ./hackbench 100
> 
> 1st run:  23.254				   16.603	
> 2nd run: 23.481					  16.478
> 3rd run:  23.790				   16.387
> 
> ./hackbench 130
> 
> 1st run:  33.395				   21.731	
> 2nd run: 32.652					  21.821
> 3rd run:  32.517				   21.698
> 
> ./hackbench 150
> 
> 1st run:  89.100				   47.862	
> 2nd run: 39.308					  25.121
> 3rd run:  90.157				   25.125
> 
> It seems to me 2.6.12-rc faster than 2.6.12-rc6-RT-V0.7.48-0. It is
> normal? 

Depends on the configuration options. If you used the same configs they 
should be the same but if you activated PREEMPT_RT it should slow things
down due to the overhead of irq-threading and  mutexes instead of
spinlocks - although I am surprised the factor is _that_ big!

Esben

> 
> thanks,
> 
> -
> mg
> 

Re: hackbench: 2.6.12-rc6 vs. 2.6.12-rc6-RT-V0.7.48-06

[Ingo Molnar]

* Esben Nielsen <simlo@phys.au.dk> wrote:

> Depends on the configuration options. If you used the same configs 
> they should be the same but if you activated PREEMPT_RT it should slow 
> things down due to the overhead of irq-threading and mutexes instead 
> of spinlocks - although I am surprised the factor is _that_ big!

one has to make sure the default debug options are disabled.  
(DEADLOCK_DETECT, PREEMPT_DEBUG, etc.) But even then there will be 
considerable overhead. Until recently, the overhead used to be even 
bigger. Note that hackbench is pretty much the worst-case for 
PREEMPT_RT: lots of locking and scheduling done and we compare the UP 
kernel to the SMP kernel in essence. (in fact PREEMPT_RT's locking is 
currently higher overhead than SMP)

	Ingo