Latency test at what load?

Hello everyone!

I am relatively new to the world of LinuxCNC and I am trying to make my first machine. I did all BIOS optimization and isolcpus= in grub. My latency turned out to be quite good Base thread = 24 uS; Servo thread = 15 uS. These results were after two hour of running time while about 25 glxgears were running in the background, which I sometimes dragged and resized. I would think that this is an adequate load, but if I really load the machine with, say, multiple libreoffice, multiple firefox, 100 glxgears, etc. then my delay went up to 80 uS on each thread.

My question would be, what would be a real ideal load that simulates the load of the linuxcnc program on the system? Because everywhere they say that I should load the machine as much as I can, but in fact I can even load it so much that my delay goes to 200, if I open, say, 250 glxgears and several libreoffice and browser tabs. But that doesn't seem like a realistic load to me.

My machine: acer q35t-am motherboard, e4600 cpu, integrated gpu. Cnc is on paralell brakeout board. 

If what i wrote is just simply bad latency would it help if I put a dedicated video card, or maybe replace the processor with an e7500 (I have one somewhere laying around)?

You are not actually trying to measure load on the machine. You are measuring the jitter between invocations of the servo and base thread.
So your servo thread is going to be 1 ms +/ 15 usec jitter
In some of the kernel traces we did, the servo thread took about 0.2 ms to run and the thread slept for about 0.8 ms. So latency would not be an issue unless it was high enough for the servo thread to run late. Worst case, the jitter will be 2x 15 usec (start 15 usec late and finish 15 usec early). So your servo thread is gunna be perfect! 250 glxgears is very excessive.

The same applies for the base thread. I have never used a parallel port so I will let others tell you how to interpret that one.