i5-8250U laptop for Linuxcnc with ethernet

PCW wrote:

Note that as long as you can reliably meet the 1 ms servo thread deadline
with a few 100 usec to spare, there is no advantage to futzing around to get
the minimum latency as it simply does not matter (at least with most hardware interfaces)

i retweak the kernel,, and running the test again with 10glxgear
the servo i m using is delta A3 and it have 3.1khz bandwidth,,, is this should match the servo thread or its not related

www.deltaww.com/Products/CategoryListT1....&Name=ASDA-A3+Series

No, assuming either position or velocity control in the drive, the servo thread
only need be fast enough to allow velocity ramp up and down with enough
steps to be smooth (this is acceleration dependent)

so when using high response servo.. i need 2khz servo thread.

i will run them in Pulse/Step,, MAX 4MHZ

the driver it self have filtering, and auto tuning, also S curve

It depends on the system acceleration and accuracy requirements.

1 KHz is usually fine. You can calculate the chord error
(caused by linear interpolation between sample period waypoints)
and see if its significant with your machine motions constraints:

Chord error polygon vs circle = R (1-cos(180/N))

R = radius
N = sides

R = Minimum Radius= V^2/A
N = Sides = T_Circle/T_Sample

this result, is it ok for external motion controller?

in another word,, am i going to gain anything if the latency is lower?

That should be fine for a 1 KHz servo thread. You can also get some idea of the network
stack latency by pinging the Ethernet card and checking the min and max ping times.

The first ping will likely be longer since the host needs to acquire the remote cards
MAC address via an ARP transaction.

about 0.22ms

If the laptop has a Intel MAC you may be able to improve this by disabling IRQ coalescing:

sudo ethtool -C edevice rx-usecs 0

where edevice is your Ethernet device name (eth1, eth0, eno1 etc etc)