1 |
On Tuesday 19 August 2008 15:02:34 Håkon Alstadheim wrote: |
2 |
> Top reports ~70% idle, while at the same time the topmost couple of |
3 |
> processes are reported as using >70%CPU. Is there anything I could use |
4 |
> that reports more sensible values ? |
5 |
> |
6 |
> I'm running the machine for multi-media-use, and I would like to make |
7 |
> sure that I tune the media-programs to leave sufficcient cpu to handle |
8 |
> the odd house-keeping task, while at the same time doing as much |
9 |
> post-processing for image and sound quality as possible. Is there a way |
10 |
> to get top to make sense, or are there other tools you good people would |
11 |
> recommend ? |
12 |
|
13 |
Let the kernel do what it does best - scheduling. You stay away from this as |
14 |
this is the one thing you do worst. Unless you have some weird workload it is |
15 |
highly unlikely that you will even remotely approach the kernel's choices for |
16 |
scheduler efficiency, much less better them. |
17 |
|
18 |
The kernel is designed to give every process a fair shot at running, this is a |
19 |
process that requires millions of decisions a second. |
20 |
|
21 |
The reason that top's output does not add up is that top (plus free and most |
22 |
of the contents of /proc as well) is basically lying through it's teeth. All |
23 |
these tools read the various files in /proc to get their data. By the time top |
24 |
has read the last file it wants, the ones before it have changed many |
25 |
thousands of times over. So basically you get a snapshot view of what the |
26 |
system is averaging, and you are not interested in the exact number you see on |
27 |
the screen (because you can't trust it), but you are interested in the general |
28 |
trend over time (as you can trust that) |
29 |
|
30 |
-- |
31 |
alan dot mckinnon at gmail dot com |