>another data point: pre8 causes stalls in a SCHED_FIFO thread of up to
>600msec! i haven't ever seen this with any previous kernel.
>
>--p
That is really bad:
(notice that Paul is running on a SCSI box)
On my box with EIDE disk (DMA=on),
I was unable to get more than 130ms latencies with a 2.2.9 kernel.
(And as you know I am stressing the box quite a bit with
comdined load, but mainly large disk I/O transfers).
Therefore there is really something wrong.
But it is interesting that 2.2.15UP + lowlatency-2.2.15 seems to perform
ok to me (when running a normal latencytest session I get <3ms latency peaks)
( no diagrams posted yet , :-( )
But that disk performance drop even in 2.2.x series worries me quite a bit,
because harddisk recording is very sensitive to disk I/O performance,
especially because the access pattern is non linear + issuing I/Os to
several files simultaneoussly.
I think I will have to modify my "hdrbench" benchmark in order to make it
simpler to use: you simply specify the file size and it generates all tracks
and files for you, and then begins testing.
(Plus, I have already included some changes (not published to make it work
on audio-less boxes), because a soundcard doesn't seem to count as
standard equipement among kernel developers :-) )
BTW: in order to compile 2.2.15 in UP mode I had to set smp_num_cpus
(undefined symbol) manually to 1
in include/linux/kernel_stat.h in the kstat_irqs function
extern inline int kstat_irqs (int irq)
{
int i, sum=0;
for (i = 0 ; i < smp_num_cpus ; i++)
sum += kstat.irqs[cpu_logical_map(i)][irq];
return sum;
}
strange ....
Benno.
-- Benno Senoner E-Mail: sbenno@gardena.net Linux low-latency audio / scheduling latency benchmarks http://www.gardena.net/benno/linux/audio- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon May 15 2000 - 21:00:25 EST