Skip to contentSkip to conversation

SMP stall on process creation

DISCUSSION Daniel Jacobowitz 2 participants 4 messages Dec 16, 2000

lore
Share & tools
lore
#0 lore raw

I've been using a dual G4/500 with the linuxppc_2_5 tree for a couple weeks now, and having all sorts of strange problems. While trying to debug one (console corruption on select, and sometimes on boot), I discovered something very interesting.

I ran, in zsh, this command:
% time (for i in {1..100}; do /bin/echo -n; done)

on several machines. On an old 604e/200:
0.52s user 1.25s system 100% cpu 1.759 total

On a dual-450 celeron system:
0.15s user 0.18s system 62% cpu 0.530 system

But on the G4:
0.07s user 6.24s system 100% cpu 6.302 total

This obviously slows shell scripts to a crawl - forking or execing is being severely penalized. I'm also a little startled by the 100% cpu - this is a multiprocessor system, that loop should NOT have been throttling both processors.

Does anyone have any ideas? I'm going to go play with 2.2 SMP and 2.4 UP for comparison.

Dan

/--------------------------------\ /--------------------------------\

|       Daniel Jacobowitz        |__|        SCS Class of 2002       |
|   Debian GNU/Linux Developer    __    Carnegie Mellon University   |
|         [email protected]         |  |       [email protected]      |
\--------------------------------/  \--------------------------------/

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

#1 lore raw
Quoted 17 lines · Daniel Jacobowitz

I've been using a dual G4/500 with the linuxppc_2_5 tree for a couple weeks now, and having all sorts of strange problems. While trying to debug one (console corruption on select, and sometimes on boot), I discovered something very interesting.

I ran, in zsh, this command:
% time (for i in {1..100}; do /bin/echo -n; done)

on several machines. On an old 604e/200:
0.52s user 1.25s system 100% cpu 1.759 total

On a dual-450 celeron system:
0.15s user 0.18s system 62% cpu 0.530 system

But on the G4:
0.07s user 6.24s system 100% cpu 6.302 total

Quoted 7 lines

This obviously slows shell scripts to a crawl - forking or execing is being severely penalized. I'm also a little startled by the 100% cpu - this is a multiprocessor system, that loop should NOT have been throttling both processors.

Does anyone have any ideas? I'm going to go play with 2.2 SMP and 2.4 UP for comparison.

[...]

Just ran this on my 9600/200mp, 2.4.0-test11 (linux-pmac-devel, not terribly current). time says:
0.84s user 2.63s system 100% cpu 3.466 total

I bumped the count to 1000, and ran top to verify. Top said (roughly) 16% user (zsh, top, X), 40% system, 44% nice (seti@home, niced 19), which is about what it should say with only one CPU throttled.

Not as zippy as it could be, but definatley not hammering both CPUs.

Cheers - Tony 'Nicoya' Mantler :)

--
Tony "Nicoya" Mantler - Renaissance Nerd Extraordinaire - [email protected]
Winnipeg, Manitoba, Canada -- http://nicoya.feline.pp.se/

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

#2 lore raw
Quoted 9 lines · Tony Mantler
Quoted 8 lines · Daniel Jacobowitz

I've been using a dual G4/500 with the linuxppc_2_5 tree for a couple weeks now, and having all sorts of strange problems. While trying to debug one (console corruption on select, and sometimes on boot), I discovered something very interesting.

I ran, in zsh, this command:
% time (for i in {1..100}; do /bin/echo -n; done)

Quoted 6 lines
Quoted 2 lines

on several machines. On an old 604e/200:
0.52s user 1.25s system 100% cpu 1.759 total

Just ran this on my 9600/200mp, 2.4.0-test11 (linux-pmac-devel, not terribly current). time says:
0.84s user 2.63s system 100% cpu 3.466 total

That's kind of upsetting right there :)

If I can figure out how to do profiling I'll try to track this down.

Dan

/--------------------------------\ /--------------------------------\

|       Daniel Jacobowitz        |__|        SCS Class of 2002       |
|   Debian GNU/Linux Developer    __    Carnegie Mellon University   |
|         [email protected]         |  |       [email protected]      |
\--------------------------------/  \--------------------------------/

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

#3 lore raw
In reply to Daniel Jacobowitz
Quoted 5 lines · Daniel Jacobowitz

I've been using a dual G4/500 with the linuxppc_2_5 tree for a couple weeks now, and having all sorts of strange problems. While trying to debug one (console corruption on select, and sometimes on boot), I discovered something very interesting.

Which was VGA console stuff, by the way. I'm trying to get a second opinion from linux-fbdev before touching the code though.

Quoted 11 lines

I ran, in zsh, this command:
% time (for i in {1..100}; do /bin/echo -n; done)

on several machines. On an old 604e/200:
0.52s user 1.25s system 100% cpu 1.759 total

On a dual-450 celeron system:
0.15s user 0.18s system 62% cpu 0.530 system

But on the G4:
0.07s user 6.24s system 100% cpu 6.302 total

And with test13:

    0.16s user 0.53s system 101% cpu 0.681 total

Much better. Whatever it was seems to have wandered away again. Note that we're still a solid thirty percent slower than x86 at this - I'll try to work out why some other time.

Dan

/--------------------------------\ /--------------------------------\

|       Daniel Jacobowitz        |__|        SCS Class of 2002       |
|   Debian GNU/Linux Developer    __    Carnegie Mellon University   |
|         [email protected]         |  |       [email protected]      |
\--------------------------------/  \--------------------------------/

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/