From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <4D430050.9070401@domain.hid> Date: Fri, 28 Jan 2011 18:43:44 +0100 From: Gilles Chanteperdrix MIME-Version: 1.0 References: <22672536.41296203668390.JavaMail.SYSTEM@PC-MRINALDI> In-Reply-To: <22672536.41296203668390.JavaMail.SYSTEM@PC-MRINALDI> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Subject: Re: [Xenomai-help] Problems with rt_task_create and rt_task_join List-Id: Help regarding installation and common use of Xenomai List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Michel Rinaldi Cc: xenomai@xenomai.org Michel Rinaldi wrote: >> Posted by Gilles Chanteperdrix on January 27, 2011 - 14:36: > >> Strange, here, the LTP test takes only around 2 hours. Just to be >> sure: you also generated disk and network activity in parallel to >> LTP and latency? If yes, then this system has no issue with >> Xenomai, and the problem you have with your application is in the >> application itself. > > I launch runltp command with -t flag (which specifies test duration) > to run LTP tests for one day. I cannot run nettests and disk I/O > tests from ltp battery because they require additional features > (former requires another linux PC with ltp installed, latter requires > floppy and CD player which aren't present on my pc). So, I create a > script that downloads via network a larger (100Mb) file, and create, > copy and delete some files on hard disk, all in a while loop, to > generate network and disk I/O traffic. I re-run ltp and latency test > running for 15 hours in parallel with this script also. I obtain: > > RTD| 1.081| 8.000| 47.249| 0| 0| 14:58:34/14:58:34 > > Now I can assert that my xenomai works grat, I think. And then, > problems are in my application. > > The only thing that looks me strange is this: with > ltp+my_I/O_stress_script+latency tests, I obtain a minimum latency > value (1.081) that is more less than ltp+latency only tests (6.182). > So, lower latency was obtained with higher system load. Maybe > there's some CPU frequency reductions in act during ltp+latency that > isn't active during ltp+latency+my_I/O_stress_script, also if in my > kernel configuration CPU frequency stepping was disabled? Or what > could be happened? The minimum path is a question of pure luck. If you are lucky that everything needed to go from the interrupt to the user-space latency task is in cache, this path takes a very short time. If you have one process which occupies cpu for long periods of times without too much impact on cache, chances are that you will get a lower minimum latency than if you have many processes switching and thrashing the cache all the time. -- Gilles.