[lug] sluggish machine

[Davide Del Vento]

Thanks everybody for the help.
I'm replying to everything in just a consolidated email.

> 1. RAM usage (is it swapping like mad?)
No, or better, I know how it sounds and I'm not complaining when it is
happening (after all if I need more than 2GB from a laptop I cannot
complain... In these cases I should use the 128GB per node we have in
bluefire, not the laptop...)

> 2. I'd wonder about a failing hard drive
> (anything in /var/log/messages that looks suspicious?)
No, but I could be overlooking something. Any keyword to grep?

> Next I'd grab a CD with memtest86 on it (CentOS install CD, for example
> has it... you just type memtest86 at the CD boot prompt)
Tried my ubuntu memtest86 availabe on the hard drive (this machine
doesn't have any CD reader)
I tried it a few times in the past, but never for so long. It took 14
hours to arrive at the 6th test (of 10), with 48% of test successfully
passed and no errors, so far. Notable data: 2038MB of memory, going at
1.8GB/s and ECC is off (I noted the whole page, but I'm not retyping
it here, should I?)

> Sounds more like a heat issue.  Check that the fan is running and that
> the machine isn't hot to the touch.  Throttling the cpu is one way the
> machine tries to protect itself.

The fan sounds ok, but the heat sink might have been detached or
something (it's a laptop, vibrations are always an issue....)
After your advice, I installed the CPU-frequency-scaling and Sensors
applet. I'm running 70-80 degrees (Celsius), since I installed the
temperature monitor, but the laptop is colder than the maximum
skin-feeling I remember.

I did my homework to know what's the normal temperature for my U7600
and I found what follows:

http://download.intel.com/design/mobile/datashts/31674505.pdf (pag 79)
indicates a 100 degree Celsium maximum operating temperature (of the
chip, non of the case neither of the air). The shut-down temperature
is 125 degrees (many other pages in the same PDF).
Same numbers in this unofficial, but better readable table:
http://www.cpu-world.com/CPUs/Core_2/Intel-Core 2 Duo Mobile U7600
LE80537UE0092M.html

So currently I'm running 20 degrees (40 F) lower than the maximum. The
skin feeling I remember didn't was this hotter, but....

... so far I noticed that the CPU-frequency sometimes doesn't scale up
quickly, and I get a little slower performance (nothing, compared to
what I was telling in my previous email, but a clue): sometimes it
sticks at 800MHz instead of scaling up to 1.2GHz.

> One way to isolate the problem to the web is to just close firefox and
> see if the problem goes away.
I'd already tried this BEFORE you suggested, and this is what I found.
I didn't have any chance to re-try it later (with the above-mentioned
applets) because it didn't happen again.
Anyway, when I closed firefox (very difficult, I had to kill it on a
non-responsive system), the machine recovered, but not quickly enough.
I mean, after you free ~200MB RAM and ~ 50% CPU usage, the system is
clearly more responsive, but still slugglish. Only after lot of time
it went back to "like new" (I didn't reboot just after because I had
many ssh sessions open that I didn't want to lose).

> You can also try to enable NoScript for firefox and see if that catches
> which page is doing evil things, if any.
This is a wonderful tool! I didn't know about it. Thanks for
suggesting, I'll keep it always running now. I need to enable almost
all of the websites I use, but I'm blocking everything I google and
everything I don't explicitly need, great! I highly recommend it to
everybody, even if currently you don't have any problem!

> i'd bet you 100% it's your browser, especially if you tab browse a lot and
> leave the browser running all day long.
I often do. But as I wrote, it didn't seem to be the whole story. Indeed....

> as someone else suggested, just kill the browser.  that'll be a quick test
> if you're performance comes back.
... the performance didn't come back right after.

So far, I think what was happening is the follows:

a) the CPU was slightly overheat (possibly triggered by that heavy
python script I mentioned)
b) the CPU-frequency downscaled (maybe due to a? I have to do some
homework on that and see how does it work)
c) something in firefox was eating more than 50% CPU-time and lot of memory
d) the rest of the apps eating the rest of the resources
e) maybe I had also bad memory fragmentation, possibly caused by
firefox and the python script

c (and maybe also d) were aggravating a

This would explain the observed behavior (including the slow "healing"
after killing firefox, due to cooling down, that is surely much slower
than freeing memory).
I will try to trigger it again, leaving firefox on for a day with a
lot of possible "evil" tabs and NoScript disabled. Then I'll run the
heavy python script. If I will reproduce the heavy slow-down, the
CPU-frequency and temperature applet will give me more info than just
the system monitor applet saying 100% CPU-usage (this will happen at
the beginning of June, until then I really will not have time to
play...)

Currently, I'm loving NoScript and I'm avoid the python script (even
if I need it, but I need other things most), so this might explain why
it's not happening.

I will only need to figure out if there is a "trivial" hw problem such
a wobble heat sink. I'll keep you posted!

Thanks again everybody and let me know if you have any additional idea!
;Dav

On Tue, 19 May 2009 16:09:56 -0600, "Davide Del Vento"
<davide.del.vento at gmail.com> said:
> Folks,
> I'm having a weird problem. My laptop is slowing down often, up to a
> point where it becomes unusable. Then it return faster as it was, or
> sometimes (like when I wrote the first draft of this email) still very
> slow, but usable - I am typing about two words forward the ones it is
> showing on the screen.
>
> *********************
> The facts:
> The machine becomes incredibly slow, system monitor reports 100% CPU
> usage, and many windows (most notably firefox) become irresponsive and
> grayed-out.
>
> *********************
> The clues:
> System monitor (as well as top) does not report a single process that
> is hogging all the resources, but it might not able to refresh often
> enough. The top-six for memory and CPU are firefox, xorg, evolution,
> thunderbird, system monitor itself and (oddly?) geyes. The rest are in
> the single digit (or even 0.something) range.
>
> The obvious suspects are javascript in the large number of pages I
> keep active in firefox (I need them, don't suggest to not use them,
> they are almost all work related, besides gmail for chatting with
> family and friends - maybe chrome would have been better to pinpoint
> the guilty site, but we don't have it for linux)
> On a similar system (older hw, same OS, maybe slightly different
> packets installed - oddly less carefully picked, like skype) they the
> same websites don't create this mess! So it's not (entirely) web's
> fault.
>
> I did deleted my .mozilla directory and let firefox re-create it
> again. Painful (autocompletion, history, settings lost, etc) and
> useless: the problem si exactly the same. I unistalled  firebug (but
> it was disabled...), I uninstalled completely firefox-3.0 and all
> related packages (gnome-support, ubufox, etc), did apt-get autoclean
> and autoremove. After reinstalling (firebug excluded) the problem is
> still here (less frequent, maybe...)
>
> RootKit might be a possibility, but nmap didn't find anything, and
> rkhunter, unhide and chkrootkit didn't find anything "too strange"
> (well, it did find something, but many people online say that are
> false positive - or are they the RootKit authors? - and then a true
> positive might hide in the list). And I trust more nmap than
> rkhunter, unhide and chkrootkit.
>
> Fragmentation may contribute to the issue, since everything started
> when I had to create a couple of multi-gigabyte files (suddenly
> raising my / usage from ~30% to ~70%) and I had to read these files
> with my own python scripts to do some data-analysis. During that time
> the system was REALLY sluggish and started to swap, but at the end the
> process was successful. But fragmentation does not explain the 100%
> CPU usage (which files should I check with filefrag -v?)
>
> *********************
> The system:
> Ubuntu Hardy 8.04.2 - fully kept updated
> Linux-2.6.24-24-generic  i686
> Gnome 2.22.3
>
> Dell Dual Core U7600 @ 1.2GHz
> 2GiB RAM
>
> Usually used with Dell dock-station and external keyboard, mouse and
> monitor.
>
> *********************
>
> Do you have any suggestion? Even if not, thanks for reading this long
> message.
>
> Bye,
> ;Dav