Fusion and RT Rreempt [Was: RTAI/fusion 0.7.2]
rpm at xenomai.org
Thu Jun 2 09:06:37 CEST 2005
Max Krasnyansky wrote:
> Gilles Chanteperdrix wrote:
>> Max Krasnyansky wrote:
>> > Yeah, but result from my app does not look very good, does it ?
>> > I mean 36.8 usec from kernel task to user task. There is also about
>> 5 usec
>> > worst case interrupt dispatch latency (measured long time ago with
>> > So the total is more than 40 usec to propagate interrupt into the
>> > Remember when you and Paolo said that you're seeing 15 usec worst case
>> > latency. I said that it's with the timer compensation and stuff,
>> > interrupt response time is much worse.
>> Timer compensation is less than 10us. The rest of the difference
>> could conceivably come from the fact that the latency test has a 100us
>> period whereas your external interrupt is triggered every 4ms, which
>> would cause disastrous cache effect.
>> What happens if you run the latency test with a 4ms period?
> Quick test (latency -p 4000 + network load + IDE load) shows
> RTS| 2249| 7150| 23975| 0| 00:04:15/00:04:15
> So you are right. It seems like it's cache related.
> Here is something even more interesting.
> I just ran the same test on the IBM Z-Pro machine:
> model name : Intel(R) Xeon(TM) CPU 3.60GHz
> stepping : 1
> cpu MHz : 3601.003
> cache size : 1024 KB
> 00:00.0 Host bridge: Intel Corp. E7525 Memory Controller Hub (rev 0c)
> 00:02.0 PCI bridge: Intel Corp. E7525/E7520/E7320 PCI Express Port A
> (rev 0c)
> 00:03.0 PCI bridge: Intel Corp. E7525/E7520/E7320 PCI Express Port A1
> (rev 0c)
> 00:04.0 PCI bridge: Intel Corp. E7525/E7520 PCI Express Port B (rev 0c)
> 01:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5721
> Gigabit Ethernet PCI Express (rev 11)
> 2.6.11, adeos r10, fusion CVS.
> latency -p 100 + high network load
> RTS| 343| 13153| 51667| 0| 00:02:30/00:02:30
> latency -p 4000 + high network load
> RTS| 1262| 780892| 1724248| 0| 00:00:19/00:00:19
> How about that :) ? 1.7 milliseconds. Wow.
> This is totally reproduceable btw. I played with the BIOS settings and
> makes no difference. MSI workaround is enabled but makes no difference.
If you look at the average figure, it says that the pathological latency
is almost continuously happening (> 780 us), so the problem is likely
related to this specific hw there. It would be interesting to know
whether running a kernel with IO-APIC and ACPI disabled exhibits the
> I was wondering what kind of numbers are you guys seeing on IA64 machines ?
4-way (1.3Ghz) Bull Novascale servers currently give ~50 us worst-case
for an earlier version of fusion used by the HYADES project that
contributed this port. UP gives ~30 us.
> RTAI mailing list
> RTAI at rtai.org
More information about the Rtai