waht's the formual for system load

[zs_hzh]

what's aspect load value should concern on ?

content switch,system call?cpu usage?what's the accurate formual would like?and why high load with more idle time !

[Steve Steel]

Hi


What os are you on

Look at


PHKL_33336:

The commands top(1) and uptime(1) may show high load averages on a 100% idle UP system.


Steve Steel

[zs_hzh]

thanks steve ,so the high load was cuased by long run queue in cpu ,but do i need to add cpu for handling those process or ,just install this patch?

CPU LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS
0 4.70 10.8% 7.1% 13.6% 68.5% 0.0% 0.0% 0.0% 0.0%
1 5.65 7.7% 2.0% 12.2% 78.2% 0.0% 0.0% 0.0% 0.0%
--- ---- ----- ----- ----- ----- ----- ----- ----- -----
avg 5.17 9.3% 4.5% 12.8% 73.4% 0.0% 0.0% 0.0% 0.0%

nbnkscs1:/ (105) root% vmstat 2 10
procs memory page faults cpu
r b w avm free re at pi po fr de sr in sy cs us sy id
15 0 0 157598 175023 136 23 0 2 38 0 0 900 9535 1710 22 9 69
10 0 0 160258 176878 185 49 2 1 29 0 0 918 11234 2257 85 12 3
10 0 0 160258 171182 217 38 1 1 24 0 0 821 11780 2231 86 13 1
10 0 0 191845 169618 154 33 1 0 24 0 0 840 10128 2228 63 13 24
10 0 0 191845 170694 128 24 0 4 15 0 0 857 9209 1953 31 6 62
10 0 0 191845 169569 166 15 0 5 10 0 0 960 11120 2266 38 17 45
14 0 0 193666 169567 147 16 0 3 12 0 0 861 10430 2109 24 7 69
14 0 0 193666 170552 145 17 0 2 14 0 0 842 9587 1912 16 12 72
2 0 0 151653 170035 147 14 0 1 12 0 0 782 9252 1705 9 11 80
2 0 0 151653 170034 136 8 0 1 7 0 0 825 9027 1712 25 8 68


Defect Description:
PHKL_33336:
( SR:8606359621 CR:JAGaf20317 )
The system and per-processor load average computations use
the absolute load on run queues at the time of computation.
This makes the computed load averages susceptible to daemon
threads becoming "ready to run" at regular intervals that
coincide with computation times. The kernel daemon threads
are also not included in load calculations.

Resolution:
Maintain and use the smoothed run queues' load to compute
the system and per-processor load averages. The kernel
daemon threads are also included in the load calculations.

[Ted Buis]

There is no precise definition for a high load. Load can be high but the system can still be meeting the service levels required. I think you are better off focusing on the response times that users are really seeing and whether or not they deem it acceptable. I would use glance or sar, vmstat, top, iostat rather than lood at load.

[Steven E. Protter]

Shalom zs_hzh,

Bill Hassell once explained to me in person how to calculate system load factor and he's posted it to numerous itrc threads that are searchable.

Sadly the explanation did not stick in my pea brain molecules. It has something to do with the number of processes waiting for CPU.

You may find comparing the top outut to a more detailed utility somewhat useful.

http://www.hpux.ws/system.perf.sh

SEP