[Nottingham] Various kernel oops

[Graeme Fowler]

On Thu, 2008-05-29 at 14:29 +0100, Martin wrote:
> It's a dual core AMD: AMD Athlon64 X2 6000

Er... OK. No HT, then? I'm not a particularly hardware-focussed person
these days, but I do know that a lot of the trickery used to make
processors seem bigger, better, faster, more is just that - trickery.
And sometimes it breaks.

The comment I made about overclocking is particularly important - I
fought with a P4 box for months which kept having Java processes
segfault and die, before discovering that is was *supplied* overclocked.
Live and learn!

> I've minimized activity on that box now. However, what I have noticed is
> that:
> 
> Mem:   8190988k total,  7589768k used,   601220k free,   573092k buffers
> 
> The used inexorably climbs ever higher and rarely drops back.

It does. That's the kernel being clever and keeping as much as it can in
memory cache.

> Checking the processes from top and X has the largest memory footprint
> at 800M or so. Everything else is just a few MBytes each. Only 176 tasks
> in total. It certainly doesn't add up to 7.5 GigaBytes!
> 
> A memory leak somewhere?

Maybe...

> And so how to investigate further?

"free"

> How does top actually report "used"?

The same way "free" does, I guess, which is essentially to
process /proc/meminfo. For reference, my desktop box at work has the
output of "free" as follows:

squonk ~]$ free
       total          used       free     shared    buffers     cached
Mem:  1024680      1010724      13956          0      31260     252724
-/+ buffers/cache:  726740     297940
Swap: 2104496        26812    2077684

...the box has 1GB RAM, appears to be using almost 1GB *but* about 250MB
is in buffers or cache. You need to check against that - if your
buffers/cache are very low, but usage is high, you have something to
look at.

Is X running a WM with loads of eye candy?

Graeme