i7 vs Atom

i7 vs Atom

[Ran Shalit]

Hello,

We are using same BSP , kernel 3.18.11 rt7, with 2 different hardwares:
1. kontron board - i7 cpu
2. portwell board. - atom cpu (E3845 )


The results are totally different with the same testing.

In kontron the error is 11usec, while in portwell it is 2000usec .

The results of portwell seems more than just bad - they same
unbelievable, the error is VERY high (2msec).

I've tried to find a benchmark which can support my results, but found
no benchmark with ATOM.

Thank you for any suggestion.

Ranran

Re: i7 vs Atom

[Seokhee Han]

Hi,


On 23/05/18 06:17, Ran Shalit wrote:
> Hello,
>
> We are using same BSP , kernel 3.18.11 rt7, with 2 different hardwares:
> 1. kontron board - i7 cpu
> 2. portwell board. - atom cpu (E3845 )
> The results are totally different with the same testing.
>
> In kontron the error is 11usec, while in portwell it is 2000usec .
>
> The results of portwell seems more than just bad - they same
> unbelievable, the error is VERY high (2msec).
>
> I've tried to find a benchmark which can support my results, but found
> no benchmark with ATOM.
I am now developing RT Linux system on minnowboard Turbot Quad Core board,
which adopts ATOM E3845 series SoC, and it shows above about 2ms worst-case
latency during cyclictest. So I try to detect SMI-related, kernel unaware,
hw latency using Linux kernel's hwlat tracer and it tells me that there are
couples of latency spikes (above 1000us) on the board.

I think that most of them are SMI-related. However, because I am not familiar
with intel system, I could not find the right way yet to disable SMI. Morever,
as googling, disabling SMI on the intel system can make the system harmful,
because some actions with SMI are related with critical conditions of the system,
like critical thermal conditions.

As you can see in the below link, you can also detect hw latency with hwlat tracer.
https://www.osadl.org/Create-a-frequency-plot-from-data-of-har.bash-script-for-hwlat-plots.0.html

Seokhee
> Thank you for any suggestion.
>
> Ranran
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Re: i7 vs Atom

[Daniel Wagner]

Hi,

On 23.05.2018 06:19, Seokhee Han wrote:
>> The results of portwell seems more than just bad - they same
>> unbelievable, the error is VERY high (2msec).
>>
>> I've tried to find a benchmark which can support my results, but found
>> no benchmark with ATOM.
> I am now developing RT Linux system on minnowboard Turbot Quad Core board,
> which adopts ATOM E3845 series SoC, and it shows above about 2ms worst-case
> latency during cyclictest. So I try to detect SMI-related, kernel unaware,
> hw latency using Linux kernel's hwlat tracer and it tells me that there are
> couples of latency spikes (above 1000us) on the board.

FWIW, I played a while with a MinnowBoard too. As it turns out the
firmware provided from Intel is just no good for -rt. I ended up
building a firmware with coreboot. With coreboot all the 2-3ms spikes
were gone. Unfortunately, building your own firmware for the MinnowBoard
is far from easy though possible:

https://3mdeb.com/firmware/building-coreboot-on-minnowboard/

Thanks,
Daniel

Re: i7 vs Atom

[Seokhee Han]

On 23/05/18 18:14, Ran Shalit wrote:
>
>
> בתאריך יום ד׳, 23 במאי 2018, 07:19, מאת Seokhee Han ‏<seokhee.han@lge.com <mailto:seokhee.han@lge.com>>:
>
>     Hi,
>
>
>     On 23/05/18 06:17, Ran Shalit wrote:
>     > Hello,
>     >
>     > We are using same BSP , kernel 3.18.11 rt7, with 2 different hardwares:
>     > 1. kontron board - i7 cpu
>     > 2. portwell board. - atom cpu (E3845 )
>     > The results are totally different with the same testing.
>     >
>     > In kontron the error is 11usec, while in portwell it is 2000usec .
>     >
>     > The results of portwell seems more than just bad - they same
>     > unbelievable, the error is VERY high (2msec).
>     >
>     > I've tried to find a benchmark which can support my results, but found
>     > no benchmark with ATOM.
>     I am now developing RT Linux system on minnowboard Turbot Quad Core board,
>     which adopts ATOM E3845 series SoC, and it shows above about 2ms worst-case
>     latency during cyclictest. So I try to detect SMI-related, kernel unaware,
>     hw latency using Linux kernel's hwlat tracer and it tells me that there are
>     couples of latency spikes (above 1000us) on the board.
>
>
>
> Hi Han,
>
> I see the 1.5 to 2 msec values every time I execute this test application
> https://github.com/nma/preempt-rt/blob/master/test-rt.c
> Can you please tell me if you get similiar results with this test application?
>
> Thanks
> Ran
>
>


Hi Ran,

I've also got the similar results (under 3ms spikes) with test-rt.c on minnowboard.

Thanks,
Seokhee


>
>     I think that most of them are SMI-related. However, because I am not familiar
>     with intel system, I could not find the right way yet to disable SMI. Morever,
>     as googling, disabling SMI on the intel system can make the system harmful,
>     because some actions with SMI are related with critical conditions of the system,
>     like critical thermal conditions.
>
>     As you can see in the below link, you can also detect hw latency with hwlat tracer.
>     https://www.osadl.org/Create-a-frequency-plot-from-data-of-har.bash-script-for-hwlat-plots.0.html
>
>     Seokhee
>     > Thank you for any suggestion.
>     >
>     > Ranran
>     > --
>     > To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
>     > the body of a message to majordomo@vger.kernel.org <mailto:majordomo@vger.kernel.org>
>     > More majordomo info at  http://vger.kernel.org/majordomo-info.html
>     >
>
>

Re: i7 vs Atom

[Seokhee Han]

Hi,


On 23/05/18 23:30, Daniel Wagner wrote:
> Hi,
>
> On 23.05.2018 06:19, Seokhee Han wrote:
>>> The results of portwell seems more than just bad - they same
>>> unbelievable, the error is VERY high (2msec).
>>>
>>> I've tried to find a benchmark which can support my results, but found
>>> no benchmark with ATOM.
>> I am now developing RT Linux system on minnowboard Turbot Quad Core board,
>> which adopts ATOM E3845 series SoC, and it shows above about 2ms worst-case
>> latency during cyclictest. So I try to detect SMI-related, kernel unaware,
>> hw latency using Linux kernel's hwlat tracer and it tells me that there are
>> couples of latency spikes (above 1000us) on the board.
> FWIW, I played a while with a MinnowBoard too. As it turns out the
> firmware provided from Intel is just no good for -rt. I ended up
> building a firmware with coreboot. With coreboot all the 2-3ms spikes
> were gone. Unfortunately, building your own firmware for the MinnowBoard
> is far from easy though possible:
>
> https://3mdeb.com/firmware/building-coreboot-on-minnowboard/
>
> Thanks,
> Daniel
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

I agree with your comment. It is not so easy to build firmware and find any
information about it, because it's my first time to handle intel SoC board.

Unfortunately, I cannot access your link. I can see just "500 Internal error"
in any pages on https://3mdeb.com/, including your link.

Anyway, Thanks a lot and welcome further information.

Seokhee

Re: i7 vs Atom

[Daniel Wagner]

> Unfortunately, I cannot access your link. I can see just "500 Internal error"
> in any pages on https://3mdeb.com/, including your link.

Ah, same here, though you can get the website from the google cache:

https://webcache.googleusercontent.com/search?q=cache:0c65iwlWfYsJ:https://3mdeb.com/firmware/building-coreboot-on-minnowboard/+&cd=1&hl=de&ct=clnk&gl=de&client=firefox-b

HTH,
Daniel

Re: i7 vs Atom

[Seokhee Han]

On 25/05/18 01:07, Daniel Wagner wrote:
>> Unfortunately, I cannot access your link. I can see just "500 Internal error"
>> in any pages on https://3mdeb.com/, including your link.
> Ah, same here, though you can get the website from the google cache:
>
> https://webcache.googleusercontent.com/search?q=cache:0c65iwlWfYsJ:https://3mdeb.com/firmware/building-coreboot-on-minnowboard/+&cd=1&hl=de&ct=clnk&gl=de&client=firefox-b
>
> HTH,
> Daniel
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Thanks a lot. It will be very helpful to me.
Seokhee