From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <51E58DD1.7060909@xenomai.org> Date: Tue, 16 Jul 2013 20:15:45 +0200 From: Gilles Chanteperdrix MIME-Version: 1.0 References: <51E4F63A.7080203@163.com> <51E55AB0.8070600@siemens.com> <51E562FD.90802@axelsw.it> In-Reply-To: <51E562FD.90802@axelsw.it> Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [Xenomai] imx28 high latency List-Id: Discussions about the Xenomai project List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Roberto Bielli Cc: xenomai@xenomai.org On 07/16/2013 05:13 PM, Roberto Bielli wrote: > Hi, > > i put xenomai 2.6.2.1 with patch 'ipipe-core-3.4.6-arm-4.patch' on a > vanilla kernel 3.4.6 for a karo tx28 board (tx28 has imx28 processor > that uses standard clock management to access the timer). > I run the latency test but seems to have very high latency. > (I saw always that every 5 records the 'lat avg' is higher than previous. Well, it seems something is running every 5s and has adverse effects, thrashing cache maybe, you can use the I-pipe tracer to know exactly. > Also when i run after some seconds the command 'ls -lR /' from a telnet > console for increaseing interrupts the 'lat min' is raised to 54.200~ us ) > i read correcly the data below or i misunderstanding the behaviour ? The imx28 is an arm926, not a cortex a-something. The figures you get are a bit low actually, I would expect the worst case latency to be above 100us. > Are there some kernel features to disable ? You can try to enable FCSE. In "guaranteed" mode, you will be limited to at most 95 processes, every one of them with at most 32MB of virtual address-space, and improved worst case latency. In "best-effort" mode, you do not have the limitations, but you only improve average latency. -- Gilles.