[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
[rtl] jitter and fpu
Ciao,
Wayne has recently presented figures about the jitter, related to interrupt latencies, under RTLinux. I support his figures in the average, as an up to 4-6 us latency is granted once the timer inturrupts while Linux is in the preamble, allowed by RTLinux, of any inturrupt servicing and keeps the interrupts disabled while writing (3) masking-unmasking io instructions to the 8259 chip.
I logged the jitter for some hours by using the rough criteria of matching the period of a periodic task writing to the parallel port, 100 us, with the cpu clock. I got up to 8 us latencies and no loss of interrupts (Pentium II-233) even forcing a substantial disk activity.
The odds come up when I turn on the fpu support by just using fsave/frstor in the task switch, no fp calculation whatsoever in the RTLinux periodic task. In that case the average jitter is more substantial and interrupts are lost, up to 2 in 100000.
Since I do not want to accept the idea that just saving and restoring the fpu registers has anything to do with the disk what is happening?
This same problem has been pointed out many times also by Daniele and got no answer. But if you are stretching RTLinux to the top of its performance be careful, you are likely to meet these troubles too.
The reason I'm writing this is the search for help: is there anybody that can explain what's happening?
Paolo.
--- [rtl] ---
For more information on Real-Time Linux see:
http://rtlinux.cs.nmt.edu/