- From: Michael Van Damme <michael.vandamme_at_vub.ac.be>
- Date: Sat, 16 Oct 2004 00:25:08 +0200
I've just tested 2.6.8.1. It isn't any better, same results as 2.4.27. Michael Bernd Porr wrote: > Michael, > > which kernel do you use? David Brownell and me we added a 5ms watchdog > timer in the 2.6.x kernel which kicks in when the host controller fails > to issue an interrupt. > > The other issue might be that comedi is buffering data but I'm not aware > of that. At least my driver does no buffering at all. > > /Bernd > > www: http://www.berndporr.me.uk/ > http://www.linux-usb-daq.co.uk/ > Mobile: +44 (0)7840 340069 > Work: +44 (0)141 330 5237 > University of Glasgow > Department of Electronics & Electrical Engineering > Room 519, Rankine Building, Oakfield Avenue, > Glasgow, G12 8LT > > > Van Damme Michael wrote: > >> Bernd, >> >> the 5 ms is when I read with a comedi command and write with >> comedi_data_write. >> If I use commands to both read and write (and I copy input to output in >> the software), I achieve 1 ms resolution on the output, but it lags >> behind >> over 20 ms. If the usb host controller is to blame, it apparently >> introduces a different delay depending on which comedi output method >> is used. Seems very strange to me. >> I'm under the impression that something somewhere is buffering for >> some time >> before ouput actually begins, and that this doesn't happen with >> comedi_data_write. >> The actual delay is not always the same by the way. For different runs >> of the >> program, it varies from just over twenty to over thirty milliseconds. >> I don't >> think this is a thread issue, since my version with comedi_data_write >> uses >> exactly the same threads and thread-synchronisation. >> >> Michael >> >> >>> Michael, >>> >>> the delay might come from the USB host controller. When the driver >>> sends out the data to the dux board it's up to the host controller to >>> send it. The driver has no control over that. However, 5ms seems to >>> be too long. The packet should go out within one USB frame, thus, >>> 1ms. The board processes the data frame by frame so the delay should >>> be only one ms. If you need lower latency USB 2.0 might be an option. >>> USB 2.0 works with 125us microframes. >>> >>> /Bernd >>> >>> Michael Van Damme wrote: >>> >>>> Hi Tao, >>>> >>>> thanks for looking into it. I've changed my code as you suggested, >>>> and now I also achieve times of around 5 ms. That's stil a long way >>>> from 1 ms though. >>>> >>>> I was indeed using two commands simultaneously. I don't think the >>>> board gets reconfigured every time, because I do get 1 ms >>>> performance on both input and output simultaneously. The only >>>> problem is that the output lags by over 20 ms. If the board was >>>> reconfigured every time there was a switch between read and write >>>> then probably neither input nor output would behave smoothly (they >>>> do now, the output almost exactly matches the input, except for the >>>> delay). >>>> With the 5 ms approach, the output is only a rough approximation of >>>> what I send in. >>>> >>>> Thanks, >>>> Michael >>>> >>>> >>>> >>>> Tao Geng wrote: >>>> >>>> >>>>> Hi Michael, >>>>> >>>>> I have just looked at your code, it seems that it runs two >>>>> Comedi_command for read and write alternately. I guess, the problem >>>>> may be that, every time Command is started, it configures the >>>>> board, this cost many time. I am doing a work similar to yours with >>>>> this board, that is, controlling a biped robot walking. I use only >>>>> one Comedi_command to read sensor signals, and use >>>>> comedi_data_write to output to the board. The servo update rate is >>>>> about 5ms, which means that, in 5ms, 7 A/D channels are read, and 4 >>>>> D/A channels are written. I know, this may be not the fastest way >>>>> to do input and output with this board at the same time, but it is >>>>> fast enough for my robot. If you are interested, I can send my code >>>>> to you. And, you can look at the robot walking movies at my page, >>>>> www.cn.stir.ac.uk/~tgeng/robot Hope to be of help. >>>>> >>>>> Regards. >>>>> >>>>> Tao >>>>> >>>>> >>>> >>> >> >> >> >> _______________________________________________ >> comedi mailing list >> comedi_at_comedi.org >> https://cvs.comedi.org/cgi-bin/mailman/listinfo/comedi > > >
Received on 2004-10-15Z21:25:08