1 |
On Wednesday 14 February 2007, brullo nulla wrote: |
2 |
> > Most likely you only looked at user cpu % and neglected to list the |
3 |
> > system and niced times as well. |
4 |
> > |
5 |
> > p.s. golden rule: ps lies. top lies. free lies. |
6 |
> > Don't believe the readings they give, rather interpret them in |
7 |
> > context. |
8 |
> |
9 |
> sob. it's not the first time I hear this. What should I believe to |
10 |
> really know my system state? |
11 |
|
12 |
You should use ps, top and free of course! Just realize that they lie... |
13 |
|
14 |
Actually they don't lie, they give an average. You also need to |
15 |
understand what is going on. Without getting into the details of |
16 |
exactly what a cpu is doing, it's performing actions millions of times |
17 |
a second. In the time it takes top to update (2 seconds), the kernel |
18 |
could have started and shut down several thousand forked processes. So |
19 |
trying to measure something that changes so fast is a waste of time. |
20 |
|
21 |
Also, there's the kernels internal timer. It's set by default to 100Hz, |
22 |
which means that the kernel updates it's own counters 100 times a |
23 |
second. It's quite possible to use and discard lots of memory in that |
24 |
gap and for it to never even show up in the kernel's counters. Raising |
25 |
the timer frequency for 1000Hz for a desktop gives more accurate |
26 |
results, but they are still just an average. |
27 |
|
28 |
Same with memory - the kernel allocates it out and shares the same |
29 |
memory between several processes as it sees fit. And does what it wants |
30 |
to with swap as well. Again, trying to find out the current state of |
31 |
the system is useless as by the time you read it, it has changed 16 |
32 |
million times.... |
33 |
|
34 |
When we say that top lies, we really mean that it is giving you an |
35 |
average over a looooooong period of cpu time, and you should interpret |
36 |
it as such. It's a trend, not an instant value. |
37 |
|
38 |
alan |
39 |
|
40 |
|
41 |
|
42 |
-- |
43 |
Optimists say the glass is half full, |
44 |
Pessimists say the glass is half empty, |
45 |
Developers say wtf is the glass twice as big as it needs to be? |
46 |
|
47 |
Alan McKinnon |
48 |
alan at linuxholdings dot co dot za |
49 |
+27 82, double three seven, one nine three five |
50 |
-- |
51 |
gentoo-user@g.o mailing list |