On Sat, 2007-03-03 at 23:24 +0100, trem wrote: > Philippe Gerum wrote: > > On Mon, 2007-02-26 at 00:50 +0100, trem wrote: > > > >> Hi > >> > >> I've tried to port the 2.6.19 x86_64 adeos patch to 2.6.20. you can find > >> it here : > >> http://zarb.org/~trem/adeos-ipipe-2.6.20-x86_64-1.7-03-trem.patch > >> > > > > The good news is that moving to 2.6.20 does not seem to be too much of a > > hell. Right? > > > > > Yes, there are few conflict so it's quite easy to port xenomai to 2.6.20. > >> I can apply it with with prepare_kernel.sh (from xenomai svn), but I > >> haven't > >> tested it yet (I mean run kernel). > >> > >> I hope this patch is good enough to be used. Otherwise, I'd be pleased > >> to fix it (maybe with a bit of help). > >> > >> > > > > We are hopefully close to have a reasonably good patch for 2.6.19. Once > > we achieve this, porting to 2.6.20 should be the next step. Thanks. > > > > > I've seen ths in the Changelog : "this closes the implementation phase > of the x86_64 port." > So I suppose that the patch for 2.6.19 is finished. I've tried it and it > works fine. > The result of latency is : > > [EMAIL PROTECTED] bin]$ sudo ./latency -T 10 > == Sampling period: 100 us > == Test mode: periodic user-mode task > == All results in microseconds > warming up... > RTT| 00:00:01 (periodic user-mode task, 100 us period, priority 99) > RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat > worst > RTD| 0.760| 1.073| 3.660| 0| 0.760| > 3.660 > RTD| 0.737| 1.063| 3.012| 0| 0.737| > 3.660 > RTD| 0.741| 1.083| 3.396| 0| 0.737| > 3.660 > RTD| 0.755| 1.088| 2.344| 0| 0.737| > 3.660 > RTD| 0.754| 1.093| 3.431| 0| 0.737| > 3.660 > RTD| 0.755| 1.089| 5.119| 0| 0.737| > 5.119 > RTD| 0.724| 1.091| 3.126| 0| 0.724| > 5.119 > RTD| 0.762| 1.089| 3.171| 0| 0.724| > 5.119 > RTD| 0.751| 1.097| 5.508| 0| 0.724| > 5.508 > ---|------------|------------|------------|--------|------------------------- > RTS| 0.724| 1.085| 5.508| 0| 00:00:10/00:00:10 > > If I don't mistake, I see a max latency of 5us. Is it realistic ? >
Yes. X-related problems Paul told us about being put aside (likely MTRR issues, still to be chased), I see < 10 us under SMP test load on a 2 x dual core Opteron 285 here (test load meaning two parallel kernel compilations (-j5), and some significant network load). We still have to benchmark this port more exhaustively and aggressively (e.g. different I/O loads and severe cache trashing, for longer than 8 hours), but the foundations look sane, especially since the Xenomai/x86_64 port currently runs with a default latency calibration set to 500 ns over this. > Now, the port to 2.6.20 is open ? we can work on ? > Yep. Have fun. > > trem > > > -- Philippe. _______________________________________________ Adeos-main mailing list [email protected] https://mail.gna.org/listinfo/adeos-main
