On Mon, 2012-06-11 at 18:12 -0400, Chris Frey wrote:
> Hi Ben,
>
> Thanks for the tips. In watching the uptime on the 3.2.0 machine, I've
> noticed that it's like a wave, on an approximate cycle of a minute, sometimes
> more, where the load dips to 0.16 or so, and then quickly spikes to 0.50
> and sometimes 0.80, with very little load otherwise.
>
> Below is the top output from the machine, running kernel 3.2.0. You can see
> that the trailing load number is about 0.57. The lowest I've seen that
> number is 0.49. On the 2.6.32 kernel, after waiting about 15 minutes, it
> drops to about 0.10, and would probably go lower in time.

But there's no sign of anything in particular using that time.

[...]
> And here's the perf report during one of those upswings in load:
[...]
> # Overhead Command Shared Object Symbol
> # ........ ........... ....................... ...............................
> #
> 39.72% swapper [processor] [k] acpi_idle_do_entry
> 17.53% swapper [kernel.kallsyms] [k] read_hpet
> 16.94% swapper [kernel.kallsyms] [k] hpet_next_event.isra.6
> 9.72% swapper [kernel.kallsyms] [k] acpi_os_read_port

This seems to indicate that most of the non-idle time is spent just
waking up and preparing to sleep again. Which suggests that something
is causing far too many wake-ups.

You can look at what's causing wake-ups by running 'powertop' (from the
package of the same name). You'll need to leave the machine alone for
about 30 seconds before it accurately shows background activity.

[...]
> And again:
[...]
> # Overhead Command Shared Object Symbol
> # ........ ............... ............................... ...................................
> #
> 23.77% swapper [processor] [k] acpi_idle_do_entry
> 7.90% ksoftirqd/0 [kernel.kallsyms] [k] check_for_new_grace_period.isra.26
> 6.40% gnome-shell libc-2.13.so [.] 0x7d04c
> 5.52% NetworkManager libglib-2.0.so.0.3200.3 [.] g_hash_table_lookup
> 5.01% gnome-shell libmozjs185.so.1.0.0 [.] 0x9e85d
> 3.21% gnome-shell libgjs.so.0.0.0 [.] 0x2612d
> 2.32% gnome-shell libgobject-2.0.so.0.3200.3 [.] 0x2aba5
> 1.84% gnome-shell libxml2.so.2.8.0 [.] 0x329cf
> 1.76% kworker/0:1 [kernel.kallsyms] [k] paravirt_read_tsc
> 1.67% swapper [kernel.kallsyms] [k] read_hpet
> 1.57% gnome-shell libglib-2.0.so.0.3200.3 [.] g_strsplit
> 1.46% gnome-shell libglib-2.0.so.0.3200.3 [.] g_static_mutex_get_mutex_impl
> 1.45% gnome-shell [kernel.kallsyms] [k] find_vma
> 1.43% nm-applet libgtk-3.so.0.400.2 [.] 0x15d152
> 1.43% gnome-shell libgobject-2.0.so.0.3200.3 [.] g_signal_emit_by_name
> 1.43% gnome-shell libgirepository-1.0.so.1.0.0 [.] g_interface_info_find_method
> 1.41% gnome-shell libc-2.13.so [.] malloc
> 1.35% nm-applet libcairo.so.2.11200.2 [.] 0x4bc87
> 1.35% gnome-shell i915_dri.so [.] 0x1387d6
> 1.30% gnome-shell libcogl.so.9.1.1 [.] 0x3f5e0
> 1.28% gnome-shell libgnome-shell.so [.] st_box_layout_child_get_type
> 1.27% nm-applet [kernel.kallsyms] [k] walk_component
> 1.25% gnome-shell libclutter-1.0.so.0.1000.6 [.] clutter_paint_volume_union
> 1.23% Xorg [kernel.kallsyms] [k] hrtimer_forward
[...]

Here the behaviour is different: gnome-shell, nm-applet and various
other processes are taking up quite a bit of time.

Maybe due to background scanning of wireless networks?

Ben.

--
Ben Hutchings
Lowery's Law:
If it jams, force it. If it breaks, it needed replacing anyway.