Re: Big jitter issues every 5 seconds

[Jan Kiszka <jan.kiszka@siemens.com>]

On 20.08.21 19:14, Ivan Jiang via Xenomai wrote:
> Dear xenomai Oners:
> 
>  
> 
>        My name is Ivan and our company use xenomai 3.10 patched on Linux 4.19.94 worked on Daul core Cortex A7 800MHz CPU.
> 
>        We used this solution to build an EtherCAT Master controller.
> 
>        With the isolcpus=1 nohz=on nohz_full=1  rcu_nocbs=1 the user space latency waveforms is like below:
> 
>  
> 
>        Normally jitter is from -10~10uS every 1mS period,
> 
>        With a -30~30uS worst jitter every 200mS which made by our wifi AP enabled on our A7 Core board, this one could be disappeared if we turned off the WIFI.
> 
>        But the most unimaginable jitter is up to -55~55uS every 5S, which could not find any reason,  that we closed all the peripherals on the board except the EtherCAT.
> 
>        Whatever we turn on/off the L2 Cache or change the size of registry slots and size of system heap / private heap / shared heap.
> 
>        Our biggest problems are why WIFI could influent the Jitters and the biggest jitter every 5S is forbidden, we must get rid of them.
> 
>        Could you help us to analysis these questions.
> 

I'm afraid, 55 µs is already quite good for such a low-end system, even
with core isolation measures.

The last-level caches are typically shared (maybe lstopo can visualize
their hierachy), and the RT workload will likely not fit into core-local
L1 caches. So you cannot avoid eventual spikes already from that
perspective.

In addition, there are likely contention points on the I/O path of your
SoC - only the SoC vendor may be able to tell you where. One you
apparently found already (access to wifi vs. access to Ethernet or
something else on the RT I/O path).

Jan

-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux