Re: External: Questions about Function: epicsEventWaitWithTimeout (id, timeout)

[Niko Kivel via Tech-talk <tech-talk at aps.anl.gov>]

Hi

I noticed that you run the 3.10 rt-patched kernel. Did you do any tuning of the system? There is a guide from CERN that's pretty decent https://linux.web.cern.ch/mrg/2/Tuna_User_Guide/, and also one from RedHat https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux_for_real_time/8/html/optimizing_rhel_8_for_real_time_for_low_latency_operation/assembly_improving-latency-using-the-tuna-interface_optimizing-rhel8-for-real-time-for-low-latency-operation

To me it looks like you have a rather stable (given the small sample size) upper bound offset of 55 us. That's your baseline processing time. If you want 250 us delay, you'd need to ask for 195 us .. in this case.

As Till mentioned you have to put load on the system. Not only CPU-load, but also disk-IO and network. Common trip wires are, automount, log rotate, network manager, ... . So basically, any process that thinks of itself as being "smart". Turn them off if possible. Turn all power saving features off. The system needs to boot, assume a state and not change that state later.

To get a good hold on the performance, tests need to be run over a long time. That is days, rather than hours. You can run for a week w/o issue, then log rotate kicks in and the proverbial s**t hits the fan.

Best

Niko

---

From: Tech-talk <tech-talk-bounces at aps.anl.gov> on behalf of 김대영 via Tech-talk <tech-talk at aps.anl.gov>
Sent: Monday, May 1, 2023 11:30 PM
To: tech-talk at aps.anl.gov <tech-talk at aps.anl.gov>
Cc: 박진용/책임/CnS팀 <pjy0429 at dawonsys.com>; 박상욱/수석/SnE팀 <screenbe at dawonsys.com>
Subject: External: Questions about Function: epicsEventWaitWithTimeout (id, timeout)

 

Dear All

First of all, thank you for your help.
We are developing a multi-core program in Epics 7.0.4. 
(Scientific Linux 7.9 / kernel 3.10.0-1160.42.2.rt56.1182.el7.x86_64)

We use a function called epicsEventWaitWithTimeout(id,timeOut).
We obtained the following Time Test results.

#1 epicsEventWaitWithTimeout( pSTDev->ST_RTthread.threadEventId,  1);
     >  Result: 1000067us (MAX) / 1000014us (MIN) 

#2 epicsEventWaitWithTimeout( pSTDev->ST_RTthread.threadEventId,  0.0005);
     >  Result: 550us (MAX) / 549us (MIN) 

#3 epicsEventWaitWithTimeout( pSTDev->ST_RTthread.threadEventId,  0.0001);
     >  Result: 154us (MAX) / 112us (MIN) 

#4 epicsEventWaitWithTimeout( pSTDev->ST_RTthread.threadEventId,  0.00005);
     >  Result: 104us (MAX) / 97us (MIN) 

#5 epicsEventWaitWithTimeout( pSTDev->ST_RTthread.threadEventId,  0.00001);
     >  Result: 64.9us (MAX) / 60.5us (MIN) 

#6 epicsEventWaitWithTimeout( pSTDev->ST_RTthread.threadEventId,  0.000001);
     >  Result: 56us (MAX) / 55.9us (MIN) 
     
#7 epicsEventWaitWithTimeout( pSTDev->ST_RTthread.threadEventId,  0);
     >  Result: 54.9us (MAX) / 54.8us (MIN) 

The time measurement method is as follows.


There are some questions based on the test results.
Q1) The parameter, timeout, is a double type, indicating seconds.
     But the result is not. Something is not accurate. What could be the cause?
Q2) How do I improve time accuracy?    
Q3) Is there any other way than using epicsEventWaitWithTimeout?

 

 

Thank you and Best Regards,