Re: Stable Linux kernel 2.6 for MPC8XX

[David Jander <david.jander@protonic.nl>]

On Friday 10 March 2006 16:33, Wolfgang Denk wrote:
> > I believe most of those observations and measurements are not valid
> > anymore. Kernel 2.6 for 8xx has come a long way since this article was
> > written. It might have been true back then, but it surely isn't anymore.
>
> So did you actually run any benchmarks? Specilations on what might be
> or should be are not really helpful.

Of course I did. Otherwise I wouldn't say this.

Here's some benchmark data from nbench (sorry didn't try lmbench yet):

The same ELDK (version 3.1.1) for both kernels, running on exactly the same 
board (MPC852T 100MHz, with 32Mbyte SDRAM and 32Mbyte Flash running from NFS 
root). I removed some FPU benchmarks, as they are pretty meaningless for this 
board and take an ethernity otherwise.

Results for Kernel 2.4.25 (Denx CVS from around sept-oct or so, 2005):

TEST                : Iterations/sec.  : Old Index   : New Index
                    :                  : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT        :          30.438  :       0.78  :       0.26
STRING SORT         :          1.5842  :       0.71  :       0.11
BITFIELD            :      7.9506e+06  :       1.36  :       0.28
FP EMULATION        :           3.258  :       1.56  :       0.36
IDEA                :          108.89  :       1.67  :       0.49
HUFFMAN             :          26.281  :       0.73  :       0.23
LU DECOMPOSITION    :         0.32765  :       0.02  :       0.01
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX       : 1.052
FLOATING-POINT INDEX: 0.257


Now the results for 2.6.14 (Denx git released 2.6.14):

TEST                : Iterations/sec.  : Old Index   : New Index
                    :                  : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT        :          32.654  :       0.84  :       0.28
STRING SORT         :          1.7408  :       0.78  :       0.12
BITFIELD            :      8.3466e+06  :       1.43  :       0.30
FP EMULATION        :           3.506  :       1.68  :       0.39
IDEA                :           115.3  :       1.76  :       0.52
HUFFMAN             :          27.855  :       0.77  :       0.25
LU DECOMPOSITION    :         0.35932  :       0.02  :       0.01
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX       : 1.115
FLOATING-POINT INDEX: 0.265


I don't know why, but while everyone still says 2.6 is slower, I am 
consistently getting results that seem to prove the opposite. Why?
Is the TLB/cache stuff better optimized for 8xx in 2.6?
IMHO it is quite a difference.
Btw, I also wrote different small "speed-measurement" tools (to measure 
loop-speed, memory throughput for different block sizes, etc...) and they all 
show aproximately the same increase.
I was careful to strip both kernels of all unnecessary drivers and features 
that could hamper performance. If you wish I could try to dig up the .config 
files for you, but I am not sure I'll find them anymore (I did this when 
2.6.14 was just released).

Greetings,

-- 
David Jander
Protonic Holland.