[Xenomai-help] Interruption of sata hard driver's during real-time task running

[Xenomai-help] Interruption of sata hard driver's during real-time task running

[Donggu Kang]

Dear all


I tried to test of latnecy with a etherlab on the xenomai framework.
The etherlab is ethercat master implimentation in the Linux Evironment
and It's using normal ethernet interface card.

In my test, I have found huge latency under hard disk stress situation.
For making hard disk stress, I ran nomal linux process that is "while:
do ls -lR / > /tmp/list; done & > /dev/ null".
After than I also run xenomai task with etherlab master api.

The latest latency is above a 500us. It is unacepptable on my system.
When I run latency test tool of xenomai, the latency is under a 5us

I think the real-time task and communication of NIC is intercepted by
hard disk IRQ.

Is there any idea to prevent interruption of sata hard driver's during
real-time task run.
Thanks for any comment or suggestion regarding that problem.

Hardware
CPU : Core 2 Duo L7400 1.5ghz
ICH : ICH7
NIC : realteck 8139c ( 8139too )
HDD : SATA 2 250 Mb 5000rpm

Real Time Environment
Debian sqeeze,
kernel 2.6.37.6,
Xenomai-2.5.6,
adeos-ipipe-2.6.37.6-x86-2.9-02.patch
EtherLab-1.5 stable
Cycle time 1ms


Best regard,

Donggu Kang.

Re: [Xenomai-help] Interruption of sata hard driver's during real-time task running

[Gilles Chanteperdrix]

On 10/21/2011 09:23 AM, Donggu Kang wrote:
> Dear all
> 
> 
> I tried to test of latnecy with a etherlab on the xenomai framework.
> The etherlab is ethercat master implimentation in the Linux Evironment
> and It's using normal ethernet interface card.
> 
> In my test, I have found huge latency under hard disk stress situation.
> For making hard disk stress, I ran nomal linux process that is "while:
> do ls -lR / > /tmp/list; done & > /dev/ null".
> After than I also run xenomai task with etherlab master api.
> 
> The latest latency is above a 500us. It is unacepptable on my system.
> When I run latency test tool of xenomai, the latency is under a 5us

I am a bit at a loss here. Is the latency 5us or 500us? If the 500us
latency is measured using a tool running through normal ethernet
interface card, it can not be considered a real-time task, even if the
task doing the measurement was created using one xenomai's skins, so the
latencies you measure are linux latlencies.

> 
> I think the real-time task and communication of NIC is intercepted by
> hard disk IRQ.
> 
> Is there any idea to prevent interruption of sata hard driver's during
> real-time task run.

When a real-time task runs in primary mode, it can not be interrupted by
any Linux interrupt. However, a real-time task using linux services
(such as linux network services), it will switch to secondary mode, and
a linux irq will be able to interrupt it.

-- 
					    Gilles.

Re: [Xenomai-help] Interruption of sata hard driver's during real-time task running

[Gilles Chanteperdrix]

On 10/31/2011 12:39 PM, Donggu Kang wrote:
> Dear Gilles
> 
> I re-made xenomai task that use etherlab kernel module.
> Xenomai task is working as kernel module.
> 
> As your comment, when I call the etherlab api, the mode of real-time
> task is changed to secondary mode.
> And the real-time task could be interrupted by linux irq. right?
> 
> but, In "Xenomai's primary and secondary domains" section of "Life
> With Adeos" says, xenomai have interrupt shield function.
> So, if I turn on the interrupt shield option, latency could be more reduced.
> 
> Is it collect ?

The interrupt shield no longer exists. If you want a task to meet its
deadlines, it should remain in primary mode.

Please do not send private mails.

-- 
                                                                Gilles.