Re: Strange time consumption in multi thread CA Client program

["Sinclair, John via Tech-talk" <tech-talk at aps.anl.gov>]

Hi,

Have you done anything to ensure that each individual thread is always running on a single cpu? Otherwise calling start=clock() might be sampling the clock of a different cpu than end=clock().

John Sinclair

ORNL

---

From: Tech-talk <tech-talk-bounces at aps.anl.gov> on behalf of Mark Rivers via Tech-talk <tech-talk at aps.anl.gov>
Sent: Monday, June 26, 2023 9:31 AM
To: tech-talk <tech-talk at aps.anl.gov>; wzhy <wzhy66 at foxmail.com>
Subject: [EXTERNAL] Re: Strange time consumption in multi thread CA Client program

 

Hi,

You said:

• I can see the camonitor within base/bin/win32-x86 suspend at the same time 
  

That implies you have built for the win32-x86 architecture, which is 32-bits.

You also said:

• My hardware is i7-3770, 8G RAM. The OS is Microsoft Windows 7_64bit
  

That means your machine is 64-bit capable.  What is EPICS_HOST_ARCH set to, win32-x86 or windows-x64?  I suggest using windows-x64 and see if the problem still happens.

Mark

---

From: Tech-talk <tech-talk-bounces at aps.anl.gov> on behalf of wzhy via Tech-talk <tech-talk at aps.anl.gov>
Sent: Sunday, June 25, 2023 9:39 PM
To: tech-talk <tech-talk at aps.anl.gov>
Subject: Strange time consumption in multi thread CA Client program

 

Hello EPICS community!

I'm writing a multi thread CA client test program to find the performance of CA put with softIOC using C language. My program is like this:

preemptive_thread_handle_function()

{

ca_attach_context(pcaput_context);

for (i = 0; i < pv_cnt; i++)

{

switch (pv's dbr_type)

{

case CHAR:

ca_array_put(DBR_CHAR, 1, pCh, value);

break;

case SHORT:

ca_array_put(DBR_SHORT, 1, pCh, value);

break;

case LONG:

ca_array_put(DBR_LONG, 1, pCh, value);

break;

case FLOAT:

ca_array_put(DBR_FLOAT, 1, pCh, value);

break;

case WAVEFORM:

ca_array_put(DBR_CHAR, array_cnt, pCh, value);

break;

}

}

ca_flush_io();

}

There are 3 threads to run this, and each thread's pv_cnt is ~3000. The main program is a infinite loop to schedule all the threads.

After running for servel mininutes,strange time consumption occured during the for loop with no rules.

The time measure code is like this:

clock_t start, end;

start = clock();

for (i = 0; i < pv_cnt; i++)

{...}

ca_flush_io();

end = clock();

double dt = (double)(end - start) / CLOCKS_PER_SEC;

printf("create_pv dt=%lf\n", dt);

In most time, the dt is ~0.002s, sometimes is ~0.015s. But suddenly, it becomes a number ~5.0s and I can see the camonitor within base/bin/win32-x86 suspend at the same time clearly.

After that, it resumes ~0.002s. The 5.0s appeared 6 times in half an hour.

EPICS base version is 3.15.8(win32_x86). My hardware is i7-3770, 8G RAM. The OS is Microsoft Windows 7_64bit and the dev IDE is Microsoft Visual Studio 2017.

My test program used CPU ~25%(60s Ave) shown by the Windows Process Monitor.Before I run my test program, the softIoc.exe starts like this:

softIoc.exe -D softIoc.dbd -d test1.db -d test2.db -d test3.db

My question is: All the PVs' type and count are fixed, why CA put consumed different time? How to avoid 5.0s situation?

Any hint and help will be appreciated.

Thank you very much!