Re: [Xenomai] Heads up: I-pipe patch status on ARM

[Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>]

On 07/29/2012 11:44 AM, Paul wrote:

> On Saturday 28 July 2012, Gilles Chanteperdrix wrote:
>> It does not look that bad. However, "dohell 3600" does not do much.
>> If you want to improve the latencies, use "-p 100" instead of "-p
>> 1000" ;-)
> 
> Could try running a 'make j8 Image' in the kernel source tree. That 
> would give the system a good run for my money.. However, dohell gives a 
> baseline load that others could repeat (my excuse, and I'm sticking to 
> it ;-p )


I agree, that is the reason we provide dohell. But what I meant is that
you should use the various options to add some load (-s adds network
trafic, -m adds some I/O, -l runs the whole LTP testsuite, which gives
the kernel the opportunity to try and perturb xenomai with many
differents paths in its code, and many different kinds of workload).

> 
>> How do you get the fancy grid with gnuplot? We could add the option
>> to scripts/histo.gp. Note that I prefer to draw graphs with a log
>> scale, I find they are more meaningful, otherwise, you do not really
>> see how the graph ends.
> 
> gnuplot> set grid
> 
> The next set of plots, I'll use a log scale as you suggest.


We have been talking about this, but it would be nice to have a database
somewhere with the results of a latency runs (the file generated by
latency -g for instance), with some data attached (anything from a
simple free from string, to a complete database record with fields such
as linux version, I-pipe version, xenomai version, dohell command line),
and have a way to generate graphs from several such records. I guess
this could be done with a web application, using svg to avoid generating
the graphs on the server.

-- 
                                                                Gilles.