both better and worse interrupt latency with the "realtime" patces

From: Bo Brantén
Date: Sat Nov 13 2004 - 14:51:48 EST


I have measured the interrupt latency on a few different kernels.

The setup is a function generator connected to ACK on the printerport and an line added to the interrupt handler that toggles an data pin on it. Both signals is viewed on an oscilloscope with storage. (The computer was an old Athlon 800MHz but the results should be considered relative to each other)

Linux 2.4.27:
most often 15us.
often 10us or 15us
time distribution decreases with time, didn't see any longer
than 250us

Linux 2.6.9:
most often 15us.
else almost always 10us to 30us
didn't see any longer than 75us
This is an improvement over the 2.4.x kernel

realtime-preempt-2.6.10-rc1-mm2-V0.7.1 by Ingo Molnar:
often up to 50us
every few seconds one interrupt takes 400us to 600us

MontaVistas 4 patces from 08 Oct againts Linux-2.6.9-rc3:
most often 10us, 15us or 25us
almost always below 30us
but every few seconds an interrupt takes 600us.
Except for the sporadic long latency this patch would be the best, it collects almost all interrupts below 30us.

Bo Branten

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at