Latency Tuning Questions

More
16 May 2019 00:28 #133935 by Project_Hopeless

PCW wrote: Another thing that helps noticeably with Preempt-RT is CPU speed and cache size.
If you replace the E2180 with a E8500 you get about double the performance and
3 times the cache size (for about 6 bucks on Ebay)...

I have an E4300 only a small step up on the E2180.

I dropped the HDD in the i7 box and got better and seemingly stable numbers with minor tweaks. servo=19K, base=60K.

I'm running a historgram at this moment, wee see what that shows. I'm think I'll just stick with the i7.

I don't need extreme step rates I'm dealing with a hobby cnc. What are good target numbers to shoot for?

Attachments:

Please Log in or Create an account to join the conversation.

More
16 May 2019 01:26 #133938 by PCW
Replied by PCW on topic Latency Tuning Questions
Unless you are running a parallel port configuration, the latency does not directly affect
the step rate. All that is required is that the servo thread can run reliably at a 1 KHz or so rate, With Mesa Ethernet hardware, 10 MHz step rates are possible, completely independent of latency, but a 1 ~KHz reliable servo thread is a must.

This is because with step generator hardware, the actual steps are generated in the interface, not LinuxCNC. All LinuxCNC does is update positions and velocities (step rates) at the servo thread rate.

Please Log in or Create an account to join the conversation.

More
16 May 2019 15:31 #133983 by Project_Hopeless

PCW wrote: Unless you are running a parallel port configuration, the latency does not directly affect
the step rate. All that is required is that the servo thread can run reliably at a 1 KHz or so rate, With Mesa Ethernet hardware, 10 MHz step rates are possible, completely independent of latency, but a 1 ~KHz reliable servo thread is a must.

So <1,000,000ns on the servo thread, even half that and its 500,000 ns. So I could possibly run my E2180 Core 2 @ ~130-200K ns on a 7i96E? Or is there more to it?

How does the Linux CNC latency test or the histogram test number relate to servo thread? Is it a 1:1 correlation?

Please Log in or Create an account to join the conversation.

More
16 May 2019 17:40 - 16 May 2019 17:41 #133990 by PCW
Replied by PCW on topic Latency Tuning Questions
The E2180 may be OK but unfortunately you just have to try, as long as you do not get
real time errors from overrunning the servo thread period, its OK.

The latency test gives a hint of what the system is capable of but it does not test a
significant number of factors that come into play with actual hardware, DMA latency,
network interface latency, cache re-fill latency, hardware locking, etc etc
Last edit: 16 May 2019 17:41 by PCW.

Please Log in or Create an account to join the conversation.

Time to create page: 0.126 seconds
Powered by Kunena Forum