Hello, I'm continuing my fun with RT-Linux. I've setup an interrupt driven task which uses the 2MHz clock (82C53) of a National Instruments PCI DIO-96 board. The interrupts trigger an rt-isr which changes 64 output lines of the IO board. The plan for the final product is to use the DAQ-STC of an NI PCI-E series MIO board. The upside is this chews very little of the foreground processing on the PPro 200MHz system I'm using (2 cpus, only using one right now); it performs much better than the periodic task used previously (http://www.med.uvm.edu/~gaffney/scope.html). The periodic task pretty much locked up the machine below a 20 microsecond period, the machine is still somewhat usable in irq-mode down to 5 microseconds (it locked at 3). The downside is that there is a very large amount of jitter, about 20 microseconds on either side of the scheduled time. This only occurs around every 5-15 seconds, but it does happen. (If folks want to look at the traces then let me know and I'll put them up, but they don't look different than the periodic task in terms of jitter). To me this sounded similar to the observation below... On Wed, 7 Oct 1998, Manoj Apte wrote: > > Its not just the 8253 access time (which IS significant .. i measured it > to be about 2us-3us per write), I did the same test using the Realtime > Clock (RTC) giving me an interrupt at a regular frequency. and same kinda > spikes showup. > > The reason i got a lil more convinced onthe bottom half argument is that, > if one looks thru the irq0 handler code, at many places it does a cli/sti > (4 actually) so I tend to believe this could be the cause. Though.. the > 20us still bothers me.. since the duration of those cli's arent THAT big. > On Saturday (7-November-1998) Paolo Mantegazza noted (with regard to short duration tasks < 100 microseconds): > The situation can be improved only by changing the basics of interrupt > menagment in Linux or by getting rid of the 8259 and ISA bus as RTL has > to register Linux interrupts anyhow, even if it does not process them. > That's where most of the jitter comes. So what are others seeing in terms of jitter for _interrupt driven_ tasks? Should I be able to do better than +/-20 us? I will be trying the Mantegazza variant, but this shouldn't help for interrupt tasks, should it? Thanks for any help you can offer, -Don --- [rtl] --- To unsubscribe: echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED] ---- For more information on Real-Time Linux see: http://www.rtlinux.org/~rtlinux/
