On 03/03/2014 07:32 PM, Bruno Tunes de Mello wrote:
> Hi Gilles,
> 
> I did the tests you said to verify the system's behavior with load .
> 
> I executed some tests using commands 'stress', cpuburn for Cortex A8 and 
> dohell with LTP, for more than 20 minutes each one.
> I printed 3 results using some of these tools together: 
> http://pastebin.com/VFavnaxW
> In each test it's indicated the commands used to generate load.
> 
> It's just to confirm if it's all right.

71us may be a bit high, but have you run xeno-test and latency at the 
same time? xeno-test runs latency, so, if you also run the latency test 
separately, one latency test will perturb the results of the other.

Also, something I noted on omap4, I have no explanation, but it seems 
latency is better on cpu 0, also, running dohell with all the options 
should provide a harder load.

I use:
xeno-test -l "dohell -s 192.168.0.5 -m /mnt -l /ltp" -c 0 -p 100 -g histo

Where 192.168.0.5 has the "discard" service enabled in inetd.conf (the 
discard service is not enabled by default).

And an USB key is mounted on /mnt.

The generated "histo" files can be used with scripts/histo.gp to generate
histograms like:
http://sisyphus.hd.free.fr/~gilles/core-3.10-latencies/panda.png

Note that since the feedback from mx6 users about the L2 "write-allocate" 
bit, the panda board latency dropped down to about 25us.

-- 
                                                                Gilles.

_______________________________________________
Xenomai mailing list
[email protected]
http://www.xenomai.org/mailman/listinfo/xenomai

Reply via email to