ULE behaviour on SMP systems

From: Christoph Herrmann <C.Herrmann_at_science-computing.de>
Date: Mon, 1 Mar 2004 15:54:35 +0100 (CET)
Hi,

I'm running 5.2 current from saturday (2004/02/28, oslevel 502104)
on a dual Athlon system.

When I compiled java (whith  make HOTSPOT_BUILD_JOBS=2 ...)
I found that only one CPU was used for both compile processes.
So I testet how much time only the build oj java needs.

With SCHED_ULE top -I typically looks like that:

last pid: 13422;  load averages:  0.00,  0.07,  0.07                                         up 1+14:05:27  11:08:25
82 processes:  3 running, 79 sleeping
CPU states: 47.7% user,  0.0% nice,  2.3% system,  0.0% interrupt, 50.0% idle
Mem: 178M Active, 101M Inact, 91M Wired, 29M Cache, 60M Buf, 95M Free
Swap: 512M Total, 180K Used, 512M Free

  PID USERNAME PRI NICE   SIZE    RES STATE  C   TIME   WCPU    CPU COMMAND
13416 root     138    0 25980K 25148K CPU1   1   0:03 22.66% 22.66% cc1plus
13421 root     137    0 22540K 21736K RUN    1   0:03 21.09% 21.09% cc1plus
13392 root      76    0  1920K  1572K CPU0   0   0:00  0.00%  0.00% top

The build needs about 2:30h. time says the follwing:

9158.566u 806.682s 2:30:43.11 110.1%    -227+1704k 33789+8607io 10950pf+0w


Then I rebuild the kernel with SCHED_4BSD and did the same again.

Now both CPUs where in use (top -I):

last pid:  3745;  load averages:  1.94,  1.92,  1.42                                         up 0+00:26:38  13:46:39
55 processes:  3 running, 52 sleeping
CPU states: 95.3% user,  0.0% nice,  4.7% system,  0.0% interrupt,  0.0% idle
Mem: 98M Active, 203M Inact, 74M Wired, 14M Cache, 60M Buf, 106M Free
Swap: 512M Total, 512M Free

  PID USERNAME PRI NICE   SIZE    RES STATE  C   TIME   WCPU    CPU COMMAND
 3738 root     129    0 36160K 35680K CPU1   1   0:06 98.62% 25.59% cc1plus
 3744 root     129    0 20968K 20572K RUN    0   0:03 72.13% 10.06% cc1plus
 3739 root      -8    0  1312K  1140K piperd 1   0:00  0.19%  0.05% as
 3740 hey       96    0  1896K  1436K CPU0   0   0:00  0.00%  0.00% top

The build needs only 1:43h.

and time says:

9205.844u 892.192s 1:43:09.48 163.1%    -171+1756k 29763+8491io 10757pf+0w

I think, this is not the way it should run.

Jeff, maybe you can have a look at this?



Best regards


    Christoph &:-)
Received on Mon Mar 01 2004 - 05:54:40 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:45 UTC