Just for fun:
$cd /usr/src/linux [2.4.10-ac1+netrand+preempt+reempt-stats]
$cat /proc/latencytimes
[snip empty report]
$make -j dep
[snip long make]
$cat /proc/latencytimes
Worst 20 latency times of 2604 measured in this period.
usec cause mask start line/file address end line/fil
e
25673 reacqBKL 1 1374/sched.c c0113219 696/sched.c
22718 spin_lock 1 547/sched.c c0111644 696/sched.c
10238 spin_lock 1 547/sched.c c0111644 303/namei.c
7599 timer 5 76/softirq.c c011a437 119/softirq.
c
6832 spin_lock 0 547/sched.c c0111644 280/time.c
6731 spin_lock 1 708/open.c c01370c8 757/open.c
6174 BKL 0 59/ioctl.c c0148204 121/ioctl.c
6097 ide0 0 585/irq.c c01084ef 647/irq.c
4753 BKL 1 1462/inode.c c0178979 1470/inode.c
4008 BKL 1 1462/inode.c c0178979 143/attr.c
3961 spin_lock 0 547/sched.c c0111644 647/irq.c
3791 reacqBKL 0 1374/sched.c c0113219 1378/sched.c
3753 BKL 0 2687/buffer.c c013cbe6 696/sched.c
3519 spin_lock 0 547/sched.c c0111644 133/file_tab
le.c
2925 spin_lock 0 1308/audio.c c01e5c0b 1319/audio.c
1990 spin_lock 0 547/sched.c c0111644 172/select.c
1793 spin_lock 0 547/sched.c c0111644 700/namei.c
1500 BKL 0 301/namei.c c0143757 133/file_tab
le.c
1387 BKL 0 837/inode.c c01776d3 647/irq.c
1208 reacqBKL 0 1374/sched.c c0113219 1470/inode.c
qubes@darklands:/usr/src/linux>
-------------
For full disclosure: I use the binary only nvidia driver and have vmware2.x
modules loaded (but unused yet). This is in X with kde loaded, gkrellm
running (reported ~280 procs at height of make), xmms running. Only one skip
in play back, and the mouse jumped a bit sometimes. :)
Now, it's off to build a kernel w/o the stats, to see it that changes
anything....and it does. The sound didn't skip this time, and the proc load
"only" got upto ~200.
This is a AMD Athlon 800 w/512 Meg ram, ext3 root, reiserfs /usr on sw raid0
over two sw raid1s.
Thomas