HZ = 1000 slows down application

[akanwar at digitarchy.com]

Hi all,
 
I am expertmenting with kernel device polling on a 4.8-RELEASE system.
 
The application I am running is a traffic pumping application that sits in
an infinite while loop. At the time of this test it was doing 6Mbps in and
5Mbps out traffic. CPU usage is 40% without polling enabled, typical CPU
usage is roughly 1/3 user, 1/3 system and 1/3 interrupt. I am using the fxp
driver.

I customized my kernel with HZ=1000 and enabled polling via sysctl...CPU
usage dropped from 40% to 30%. Great so far.

But now I noticed that my application is occassionally doing slower
iterations. Average iteration time used to be 0.2 ms without polling
enabled. With the device polling changes, the average time is still around
the same, but once every few minutes the application sees iterations that
are 3.3 seconds (*seconds*, not a typo) long. 

This seems to happen as soon as I use the kernel with HZ=1000. Enabling or
disabling device polling does not seem to make any difference to this
behavior. I am trying to understand why there seem to be a few really long
iterations. Could it happen that the application does not get any CPU for
that long? Seems very counter intuitive that higher HZ should cause this. 

Could anyone shed any light on what is happening ?

Thanks,
-ansh

--------------------------------------------------------------------
mail2web - Check your email from the web at
http://mail2web.com/ .