On 07/17/2012 09:28 PM, Vladimir Mosgalin wrote:
> Hi Orion Poplawski!
>
>   On 2012.07.17 at 16:46:19 -0600, Orion Poplawski wrote next:
>
>>> That depends heavily on what version of Fedora they are (particularly Rawhide
>>> instances), and what services they are running.
>>
>> Well, I have a Fedora 17 instance running now with nothing but
>> kernel processes and the qemu-kvm process still shows using 7-8%
>> cpu.
>
>
> Run "vmstat 1 10" on both guests and compare output; look for
> differences in amount of context switches, CPU system usage and such.
>

F17 does seem to show periodic bursts of interrupts.

In one minute, F17 shows:

182  virtio0-input
17   virtio1-requests
1617 Local timer interrupts
583  Rescheduling interrupts
663  Function call interrupts
17   TLB shootdowns

SL6:
179  virtio0-input
17   virtio1-requests
849  Local timer interrupts
0    Rescheduling interrupts
0    Function call interrupts
0    TLB shootdown

So perhaps that's a clue.

> Also running powertop and comparing wakeups/sec and general output on
> guests might give some hint.
>
>

Unfortunately we have 2.0 on F17 and 1.11 on SL6 so the out put is a bit 
different.  But on SL6:

Top causes for wakeups:
   25.4% (  4.3)     <kernel core> : hrtimer_start (tick_sched_timer)
   13.6% (  2.3)       <interrupt> : virtio0-input
   13.0% (  2.2)            mysqld : hrtimer_start_range_ns (hrtimer_wakeup)
   11.8% (  2.0)          rpc.gssd : hrtimer_start_range_ns (hrtimer_wakeup)
   11.8% (  2.0)     <kernel core> : clocksource_watchdog (clocksource_watchdog)
    8.9% (  1.5)          events/0 : queue_delayed_work (delayed_work_timer_fn)
    5.9% (  1.0)             httpd : hrtimer_start_range_ns (hrtimer_wakeup)
    4.7% (  0.8)     <kernel core> : hrtimer_start_range_ns (tick_sched_timer)
    1.2% (  0.2)     <kernel core> : bdi_arm_supers_timer (sync_supers_timer_fn)
    0.6% (  0.1)     <kernel core> : start_bandwidth_timer (sched_rt_period_timer)
    0.6% (  0.1)             httpd : sys_epoll_wait (process_timeout)
    0.6% (  0.1)              hald : hrtimer_start_range_ns (hrtimer_wakeup)
    0.6% (  0.1)     <kernel core> : neigh_timer_handler (neigh_timer_handler)
    0.6% (  0.1)       flush-252:0 : schedule_timeout_interruptible 
(process_timeout)
    0.6% (  0.1)       bdi-default : bdi_forker_task (process_timeout)


Not much on F17 either:
Summary: 22.5 wakeups/second,  0.0 GPU ops/second, 0.0 VFS ops/sec and 0.4% 
CPU use

                 Usage       Events/s    Category       Description
              87.4 µs/s       6.9        Interrupt      [1] timer(softirq)
             165.6 µs/s       3.7        Timer          rh_timer_func
              57.0 µs/s       2.7        Interrupt      [3] net_rx(softirq)
              58.3 µs/s       1.9        Process        /usr/sbin/rpc.gssd
              46.7 µs/s       1.4        Timer          clocksource_watchdog
               2.1 µs/s       1.0        kWork          sync_cmos_clock
             270.1 µs/s       0.4        Process        packagekitd
             696.5 µs/s      0.20        Process        powertop
             122.4 µs/s       0.3        Process        gdbus
               3.3 µs/s       0.3        Process        [fsnotify_mark]
             568.5 µs/s      0.10        Process        /usr/lib/systemd/systemd

But there are bursts of swapper:

Summary: 16.5 wakeups/second,  0.0 GPU ops/second, 0.0 VFS ops/sec and 20.5% 
CPU use

                 Usage       Events/s    Category       Description
             144.9 ms/s      0.00        Process        swapper/3
              58.1 ms/s      0.00        Process        swapper/0
             188.5 µs/s       3.8        Timer          rh_timer_func

But the qemu-kvm cpu % is fairly steady on the host.

-- 
Orion Poplawski
Technical Manager                     303-415-9701 x222
NWRA, Boulder Office                  FAX: 303-415-9702
3380 Mitchell Lane                       [log in to unmask]
Boulder, CO 80301                   http://www.nwra.com