[ltp] Re: Periodic freeze

[Elias Oltmanns]

[Sorry for the delayed response. I'm still dealing with this problem on
=E2=80=9Ccasual observations warrant casual responses when time permits=E2=
=80=9D
priority, which is fine by my notebook but may not be the most
satisfying approach from your point of view.]

(2012-07-06) John Magolske <listmail@b79.net> wrote:
> * Elias Oltmanns <eo@nebensachen.de> [120703 17:22]:
>> (2012-07-03) John Magolske <listmail@b79.net> wrote:
>> [...]
>>=20
>> I have had occasion to observe sluggish responses too. (Un)fortunately,
>> I can't reliably reproduce this state of affairs. One thing I noticed,
>> though, is that top showed decidedly odd signs of activity. With no home
>> work on my account, the box said it was something like 80% idle (or even
>> less). Mind you, this is with no X server running at the time, or
>> anything else worth mentioning for that matter. Instead, one or two of
>> those kworker threads seemed to be anxious to make an eager impression.
>> Since this happens only every once in a while on my machine (x200s) and
>> I have to read up on domesticating those kworker threads first, I'm not
>> sure I'll get to the heart of the matter too soon.
>=20
> Thanks for writing this up, at least I know I'm not the only person
> experiencing these issues. The only solution is to plug into AC.
> Riding in on the train today, running on battery power, this display
> on this X200s was freezing a couple times a minute, if not more. Now
> that I'm plugged into AC, everything is nice & fluid

Now, this is most interesting. Here is the sequence of events as
observed just now:
1. Machine has been running flawlessly for some time with the occasional
   suspend to ram (always on AC and no X).
2. Started X and Firefox, which is when the system started to feel
   unresponsive. Sure enough, those kworker threads began showing up on
   the top ranks in top (this phrase sounds like a bad joke).
3. The sluggish response pattern remained after killing the X session
   and even persisted after another suspend / resume cycle (still
   constastly on AC).
4. Pulling the plug, the kworker threads did not rank inappropriately
   high in top anymore, but the sy% count certainly remained
   inappropriately high (still between 10% and 30%).
5. Going back to AC (still no X running), everything seems to have
   returned to normal and expected behaviour.

Unfortunately, I still have no clue as to what is really going on, I
just thought I'd let you know.

>
>> > Also, after an `aptitude upgrade` last week I've begun to hear
>> > periodic ~500ms gaps in audio while listening to music. In the past
>> > I've *never* heard the audio skip. Not sure if this is related to the
>> > display freeze-ups...who knows.
>>=20
>> This, too, is something I have observed here. By the way, it is my vague
>> impression that this only happened after I upgraded to 3.2.21 or
>> something thereabouts. But I did not really do extensive tests to verify
>> that it does not happen on 3.2.14, for example, yet.
>=20
> Do you recall which kernel you were running just prior to 3.2.21?
> I'm on Debian Sid and saw this behaviour after a recent upgrade of
> the 3.2.0-2-686-pae package (3.2.20-1).

Well, I'm running custom kernel builds and update fairly regularly.
Unfortunately, I really don't remember when I first noticed this
behaviour.

[...]
>> Can you see anything suspicious in top when you experience those
>> freezes on a supposedly idle machine?
>=20
> For me these freezes last about 3-4 seconds max, so it's hard to
> investigate too deeply. I take a look with top or htop now and then,
> but haven't really been able to see any evidence of excessive CPU
> usage.

When I notice sluggish responses, I start top and keep watching it for a
few refresh cycles. Usually something suspicious shows up well before a
minute has passed. But then our problems still might be unrelated after
all. Who knows for sure?

Regards,

Elias