You are not logged in.
Pages: 1
I have noticed htop (as well as qps and KSysGuard) shows rather strange CPU consumption (but 'top' is always OK). This is an example comparing top and htop - look at process with id 8817: http://gaydenko.com/tmp/top-01.png The process is just an http server running under JVM and loaded with ApacheBench.
Under official kernel all meters are consistent. Any ideas where to dig in?
Edit. linux-ck-corei7 is installed, i5 760 is in use.
Last edited by student975 (2011-10-31 16:35:08)
"I exist" is the best myth I know..
Offline
To try to invent something reproducible I have installed nginx http server. Again I have loaded the server with ApacheBench (ab-commands are visible at screenshots). At this screenshot the server is started with default config:
http://gaydenko.com/tmp/top-02.png
At this one I have set nginx's workers count to 8:
http://gaydenko.com/tmp/top-03.png
In fact this last screenshot is sufficient without any explanation at all - just sum up all the consuming CPU percentages (I have four cores CPU).
Again, such "jumping behavior" is demonstrated by qps and KSysGuard monitors also, not htop only.
Can anybody reproduce this jumps?
@graysky
Definitely ck-patch breaks some convention used by all the affected load monitors. I know you have contact with Con Kolivas. Can you point him to this thread? Would it be relevant?
Last edited by student975 (2011-11-01 16:22:11)
"I exist" is the best myth I know..
Offline
Disable the option "Show userland threads" and "show kernel threads" in htop options
Best Testing Repo Warning: [testing] means it can eat you hamster, catch fire and you should keep it away from children. And I'm serious here, it's not an April 1st joke.
Offline
Disable the option "Show userland threads" and "show kernel threads" in htop options
There are now "show", but there are "hide" options. I have enabled both. This screenshot corresponds to the same test as top03.png (8 worker processes and the same ab command):
http://gaydenko.com/tmp/top-04.png
'top' at this moment shows (in sum) about 380% for ab and all nginx processes (as expected).
Last edited by student975 (2011-11-01 17:56:10)
"I exist" is the best myth I know..
Offline
@student - just email him directly with your bug report or point him to this thread.
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
OK, I will.
BTW, I have tried generic built instead of cotei7 one and also have got this strange behavior.
"I exist" is the best myth I know..
Offline
Following events:
- I have contacted with Con Kolivas, he replied he don't know the way htop get the data,
- I have contacted with htop maintainer Hisham H. Muhammad. He described the way htop get the data,
- I have sent the description to Con (three days ago, still didn't receive an answer).
Also Hisham caught top's anomaly we all have missed: while sum CPU consumption is close to max, idle value in Cpu(s) string is close to 100%: http://gaydenko.com/tmp/top-01.png So, top is also affected by ck patches.
kernel 3.1 (repo-ck/linux-ck-corei7 package) also contains this issue.
"I exist" is the best myth I know..
Offline
@student - thanks for following up on this with ck. Any word from him yet?
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
@graysky,
No, there isn't.
"I exist" is the best myth I know..
Offline
Pages: 1