Re: BUG Realtime-preempt 2.6.16-rt2?

[Carlos Antunes]

On 11/1/05, Esben Nielsen <simlo@xxxxxxxxxx> wrote:
>
>
> On Tue, 1 Nov 2005, Carlos Antunes wrote:
>
> > On 11/1/05, Carlos Antunes <cmantunes@xxxxxxxxx> wrote:
> > > On 11/1/05, Esben Nielsen <simlo@xxxxxxxxxx> wrote:
> > > > On Tue, 1 Nov 2005, Carlos Antunes wrote:
> > > >
> > > > > Hi!
> > > > >
> > > > > I've been developing some code for the OpenPBX project
> > > > > (http://www.openpbx.org) and wrote a program to test how the system,
> > > > > responds when hundreds of threads are spawned. These threads run at
> > > > > high priority (SCHED_FIFO) and use clock_nanocleep with absolute
> > > > > timeouts on a 20ms loop cycle.
> > > > >
> > > > > With the stock 2.6.14 kernel, I get latencies in the order of several
> > > > > milliseconds (but less than 20ms) when running 1250 threads
> > > > > simultaneously. However, when I switch to a kernel patched with
> > > > > realtime-preempt latency increases to several hundred milliseconds in
> > > > > many cases.
> > > >
> > > > There is only one explanation:
> > > > Some of the operations (task switch, nanosleep etc.) are more expensive in
> > > > the RT kernel. Thus your 1250 threads spend 100% CPU doing what they do.
> > > > You therefore get very bad latencies.
> > > >
> > >
> > > Esben,
> > >
> > > Thanks for replying. Let me chalenge this assumption of yours, though.
> > >
> > > I just ran a test with those 1250 threads (all they do is sleep for
> > > 20ms, wake up, increment a number, and repeat the process). The CPU
> > > was 86% *IDLE* while running this. One thread took 1.3 seconds to wake
> > > up once. Do you think this is, well, normal, given how RT is supposed
> > > to operate?
> > >
> >
> > Esben,
> >
> > If, instead of SCHD_FIFO, I use SCHED_OTHER, I get max latency in the
> > order 13ms running those 1250 threads. With SCHED_FIFO (the only
> > change), I get 1.3 seconds. Makes sense to you?
> >
>
> No. I can't explain that one. You might have discovered a bug in
> SCHED_FIFO. What about SCHED_RR?
>

Same thing with SCHED_RR. It appears the realtime-preempt patch breaks
realtime scheduling! It doesn't matter if I use priority 1 or 50 or
99. I always get the same problems with SCHED_FIFO and SCHED_RR. With
SCHED_OTHER, the system is much more responsive.

How do we go about calling Ingo's attention to this?

Carlos

--
"We hold [...] that all men are created equal; that they are
endowed [...] with certain inalienable rights; that among
these are life, liberty, and the pursuit of happiness"
        -- Thomas Jefferson
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/