Re: CPU Load average VS Idle %



howa wrote:
Jean-David Beyer 寫道:

howa wrote:
using the top command, e.g.

top -d 5

the load average is : 2.59, 2.18, 1.90

However, the CPU Idle % is always 9x %

is this normal?

Insufficient data. My machine has a higher load average than yours, but I
have two hyperthreaded processors and four processes are BOINC ones that run
at nice level 19; i.e., only if nothing else wants a processor.

A high load average is not a problem unless it is "too high." For my
machine, that would be anything over 8.

Note here that my iowait is zero because there are always compute-limited
processors that can use a processor. Once in a while this misses for a
moment, but it is rare. Also note that different versions of _top_ look
slightly different, but you should be able to tell what's what.

12:28:30 up 6 days, 13:56, 4 users, load average: 4.22, 4.27, 4.32
134 processes: 128 sleeping, 5 running, 0 zombie, 1 stopped
CPU states: cpu user nice system irq softirq iowait idle
total 8.2% 80.4% 11.3% 0.0% 0.0% 0.0% 0.0%
cpu00 21.3% 65.0% 13.5% 0.0% 0.0% 0.0% 0.0%
cpu01 10.5% 79.8% 9.6% 0.0% 0.0% 0.0% 0.0%
cpu02 0.0% 86.5% 13.4% 0.0% 0.0% 0.0% 0.0%
cpu03 0.9% 90.2% 8.7% 0.0% 0.0% 0.0% 0.0%
Mem: 8208852k av, 7982092k used, 226760k free, 0k shrd, 465608k buff
3698628k actv, 3048256k in_d, 142416k in_c
Swap: 8193076k av, 0k used, 8193076k free 6108852k cached

It would be useful to know what the irq, softirq, iowait, and idle times are
for your machine. If by 9x% you mean 90% and over, it means processes cannot
use the processor because they are waiting for something other than a
processor, and IO is the most common.

more info:

20:56:22 up 250 days, 20:51, 1 user, load average: 4.49, 4.32, 4.13
102 processes: 101 sleeping, 1 running, 0 zombie, 0 stopped
CPU0 states: 5.2% user 0.4% system 0.0% nice 0.0% iowait
93.3% idle
CPU1 states: 5.2% user 1.2% system 0.0% nice 0.0% iowait
92.4% idle
CPU2 states: 8.4% user 1.2% system 0.0% nice 0.0% iowait
89.2% idle
CPU3 states: 5.0% user 0.3% system 0.0% nice 0.0% iowait
94.1% idle
Mem: 3355136k av, 3248108k used, 107028k free, 0k shrd,
111304k buff
1208084k active, 775668k inactive
Swap: 8184188k av, 2124k used, 8182064k free
1870928k cached


2 Xeon CPU, Hyperthreaded

I sure would not expect to see that. It is a lot like mine, but I would
expect you to have a lot of process time in either nice or idle, and you do not.

What does _vmstat_ show? Mine is like this at the moment, and the machine is
not doing much except four BOINC jobs (compute limited, nice 19).

$ vmstat 10
procs memory swap io system cpu
r b swpd free buff cache si so bi bo in cs us sy id wa
5 0 0 363760 472332 6326036 0 0 4 9 6 11 12 5 0 0
5 0 0 381480 472332 6326032 0 0 0 34 217 607 96 4 0 0
4 0 0 335880 472332 6326032 0 0 0 139 164 614 96 4 0 0
4 0 0 335440 472332 6326032 0 0 0 46 125 370 97 3 0 0
6 0 0 324164 472332 6326032 0 0 0 80 124 411 97 3 0 0
4 0 0 348596 472332 6326032 0 0 0 79 121 378 97 3 0 0
5 0 0 317840 472332 6326036 0 0 0 95 251 710 96 4 0 0

(Remember to not rely on the first line of output from vmstat.)

--
.~. Jean-David Beyer Registered Linux User 85642.
/V\ PGP-Key: 9A2FC99A Registered Machine 241939.
/( )\ Shrewsbury, New Jersey http://counter.li.org
^^-^^ 10:10:01 up 7 days, 11:37, 3 users, load average: 4.39, 4.29, 4.22
.



Relevant Pages

  • [GIT PULL] Scheduler updates for v2.6.36
    ... adjust when cpu_active and cpuset configurations are updated during cpu on/offlining ... Change nohz idle load balancing logic to push model ... static inline int cpuset_init ... * In the semi idle case, use the nearest busy cpu for migrating timers ...
    (Linux-Kernel)
  • Re: CPU Load average VS Idle %
    ... My machine has a higher load average than yours, ... Note here that my iowait is zero because there are always compute-limited ... CPU states: cpu user nice system irq softirq iowait idle ...
    (comp.os.linux.misc)
  • Re: [RFC] (How to) Let idle CPUs sleep
    ... turns out that if we restrict the amount of time idle cpus are ... cpu sleeps. ... * local timer ticks. ... +int idle_balance_retry ...
    (Linux-Kernel)
  • Too many timer interrupts in NO_HZ
    ... CPU in an SMP server. ... However I am interested to see CPU idle times for couple minutes on ... There are way too many timer interrupts even though the CPUs have ... entered tickless idle loop. ...
    (Linux-Kernel)
  • Re: [patch 4/7] sched: Change nohz ilb logic from pull to push model
    ... all idle CPUs indeed go into nohz mode. ... The kickee CPU does the idle load balancing on behalf of all idle CPUs ... Kicked cpu can be nearest idle core to the busy core. ...
    (Linux-Kernel)