Skip to contentSkip to conversation

Wrong calculation of load average ?

DISCUSSION Patrick Mau 2 participants 2 messages Oct 24, 2004

lore
Share & tools
lore
#0 lore raw

Hallo everyone,

I'm using a BK snapshot of linux 2.6 from today and have a strange problem with an otherwise stable SMP System. The calculated load average seems to "wrap around".

The exact kernel version is:
Linux oscar 2.6.10-2.6bk #1 SMP Sat Oct 23 16:45:50 CEST 2004

Below is the output of a looping 'cat /proc/loadavg'. The load drops to near-zero and then wraps to 1024.

I already looked at the code involved but I am unable to see how this could happen. Timer interrupts are occuring on both CPU's:

root@oscar] cat /proc/interrupts
CPU0 CPU1
17159225 17408595 IO-APIC-edge timer

The system is a dual P3 450Mhz (Asus P2B-DS), ACPI not compiled in, Preempt disabled, no regparams, but 4K stacks. 2.6.9 vanilla did not show these symptoms. System performance is not affected at all, but exim refuses to deliver mail ;)

The gcc used is:
[root@oscar] gcc --version
gcc (GCC) 3.3.5 (Debian 1:3.3.5-1)

I also tried a gcc 3.4 branch, but that didn't help either. This report is based on a kernel compiled with gcc 3.3

[root@oscar] /opt/gcc/bin/gcc --version
gcc (GCC) 3.4.3 20041017 (prerelease)

I'd appreciate any comments,
Patrick

[root@oscar] while true ; do cat /proc/loadavg ; sleep 10 ; done 112.95 1.94 2.54 1/139 25916
94.81 1.71 2.46 1/139 25918
79.47 1.49 2.38 1/139 25920
66.56 1.29 2.31 1/139 25925
55.56 1.08 2.23 1/139 25927
46.25 0.88 2.15 1/139 25929
38.59 0.74 2.09 1/139 25931
31.89 0.55 2.01 1/139 25933
26.21 0.36 1.94 1/139 25935
21.41 0.19 1.86 2/139 25937
17.35 0.02 1.79 1/139 25939 <--- here
13.92 1006.85 1.72 1/139 25941
11.01 973.53 1.64 1/139 25943
8.55 941.31 1.57 1/139 25945
6.46 910.15 1.50 1/139 25947
4.70 880.02 1.43 7/139 25949
3.28 850.89 1.37 1/139 25960
2.16 822.74 1.31 1/139 25970
1.14 795.50 1.25 1/142 25981
0.41 769.19 1.20 1/140 25983 <--- and here
941.66 743.72 1.13 1/158 26114
796.19 719.08 1.07 1/140 26164
673.09 695.25 1.01 1/140 26167
568.84 672.19 0.95 1/140 26169
480.61 649.89 0.88 1/140 26171
405.95 628.33 0.82 1/140 26173
342.77 607.48 0.76 1/140 26175
289.30 587.31 0.69 1/139 26177
244.14 567.82 0.64 1/139 26179
205.91 548.97 0.58 1/139 26181
173.48 530.73 0.52 1/139 26183
146.04 513.09 0.46 1/139 26185
122.82 496.03 0.40 1/139 26187
103.17 479.53 0.35 3/139 26189
86.77 463.63 0.30 1/139 26194
72.81 448.23 0.26 2/139 26196
60.85 433.31 0.20 1/139 26198

#1 lore raw
Quoted 3 lines · Patrick Mau

I'm using a BK snapshot of linux 2.6 from today and have a strange problem with an otherwise stable SMP System. The calculated load average seems to "wrap around".

I'm seeing the same behavior on two hosts running 2.6.9-rc4-mm1.

4K stacks, ACPI in, dual P4 Xeon with ht, 2 Gb RAM, gcc 3.3.4

Quoted 1 line

root@oscar] cat /proc/interrupts

CPU0 CPU1 CPU2 CPU3

  0:  413691994      14885  591126364          0    IO-APIC-edge  timer
  1:       1027          0          0          0    IO-APIC-edge  i8042
  8:          2          0          0          0    IO-APIC-edge  rtc
  9:          0          0          0          0   IO-APIC-level  acpi
 15:         28          0          1          0    IO-APIC-edge  ide1
 18:     186057          0    6578004          0   IO-APIC-level  megaraid
 24:  483544380     131974  554409619      23181   IO-APIC-level  eth2
NMI:          0          0          0          0 
LOC: 1004835629 1004835628 1004835628 1004835626 
ERR:          0
MIS:          0
Signature · 1 line
My other computer is your Windows box.