Q:Howto benchmark preemptible kernel ?

Q:Howto benchmark preemptible kernel ?

[szonyi calin]

Hi 
I'm have a Cyrix 486/66 with 12Megs of ram.
I'm using preemptible patches from quite some time
now.
(2.4.10- 2.4.13)
I was using both Robert Love and Andrew Morton's
patch.
I would like to do some benhmarks but there are some
issues:
1. The system is slow and has low memory so a big
benchmark is out of question (compiling the kernel
take 4 hours if i don't touch the console)
2. The benchmark must be small and adequate (patching
the kernel to make a benchmark is out of discussion)

Any ideas ?

(No i don't have money to buy a new machine, I win
200$/month, a new pc (crappy) is 350-400$ (in my
country)).

I did run (some time ago, without preemptible kernel)
some benchmarks (bonnie, and bytebench or something)
but it was no improvement (1% maybe ) whatever I have
done with hdparm or vm tweaking.
 
P.S. (for Andrew ) Will there be a patch for 2.4.14 ?


__________________________________________________
Do You Yahoo!?
Find a job, post your resume.
http://careers.yahoo.com

Re: Q:Howto benchmark preemptible kernel ?

[Andrew Morton]

szonyi calin wrote:
> 
> Hi
> I'm have a Cyrix 486/66 with 12Megs of ram.
> I'm using preemptible patches from quite some time
> now.
> (2.4.10- 2.4.13)
> I was using both Robert Love and Andrew Morton's
> patch.
> I would like to do some benhmarks but there are some
> issues:
> 1. The system is slow and has low memory so a big
> benchmark is out of question (compiling the kernel
> take 4 hours if i don't touch the console)
> 2. The benchmark must be small and adequate (patching
> the kernel to make a benchmark is out of discussion)
> 
> Any ideas ?
> 

None of these patches make any significant difference
to throughput of anything, really.

If you have a particular latency-sensitive application then
that's the thing which you should be testing with.

There's a modified version of Mark Hahn's `realfeel' app in 
http://www.uow.edu.au/~andrewm/linux/amlat.tar.gz
which I find to be a convenient way of quantitatively
determining latencies.  There are some grubby scripts
in there which create graphical output too.

 
> P.S. (for Andrew ) Will there be a patch for 2.4.14 ?

J Sloan was the first to send me an updated patch this time.
Thanks!

It needs a bit of maintenance at present - last time I gave it
a good beating (a couple of weeks ago) there were a couple of
ten millisecond blips.

-

Re: Q:Howto benchmark preemptible kernel ?

[szonyi calin]

--- Andrew Morton <akpm@zip.com.au> wrote:

> 
> None of these patches make any significant
> difference
> to throughput of anything, really.
> 

Shell scripts and related tools (i.e. make) run
faster.


> If you have a particular latency-sensitive
> application then
> that's the thing which you should be testing with.
> 

Gcc seems to be a "latency-sensitive application"
because it runs faster (but it could be the mm
improvements in kernel 2.4 -- i have't use a
non-preemptible kernel from a long time )

> There's a modified version of Mark Hahn's `realfeel'
> app in 
> http://www.uow.edu.au/~andrewm/linux/amlat.tar.gz
> which I find to be a convenient way of
> quantitatively
> determining latencies.  There are some grubby
> scripts
> in there which create graphical output too.
> 
>  

I'll give it a try. 
Thanks


__________________________________________________
Do You Yahoo!?
Find a job, post your resume.
http://careers.yahoo.com