Hi Toni,
Something appears to be artificially throttling your threads. Can you
create a thread dump? kill -QUIT pid should do it. I've not seen RH
limit threads to this extent, not without seeing system times
significantly higher.
Regards,
Kirk
Toni Menendez Lopez wrote:
1st.- CPU usage .-
12:31:54 PM CPU %user %nice %system %iowait %idle
12:31:56 PM all 28.62 0.00 0.50 0.12 70.75
12:31:56 PM 0 31.16 0.00 0.50 0.00 68.34
12:31:56 PM 1 22.11 0.00 0.50 0.00 77.39
12:31:56 PM 2 30.35 0.00 1.00 0.00 68.66
12:31:56 PM 3 30.00 0.00 0.50 1.00 68.50
12:31:56 PM CPU %user %nice %system %iowait %idle
12:31:58 PM all 31.09 0.12 0.75 0.00 68.04
12:31:58 PM 0 33.33 0.00 1.00 0.00 65.67
12:31:58 PM 1 29.70 0.50 0.00 0.00 69.80
12:31:58 PM 2 24.00 0.00 1.00 0.00 75.00
12:31:58 PM 3 38.00 0.00 0.50 0.00 61.50
12:31:58 PM CPU %user %nice %system %iowait %idle
12:32:00 PM all 29.00 0.00 0.38 0.38 70.25
12:32:00 PM 0 37.50 0.00 1.00 0.00 61.50
12:32:00 PM 1 20.60 0.00 0.50 0.00 78.89
12:32:00 PM 2 26.13 0.00 0.00 1.01 72.86
12:32:00 PM 3 31.34 0.00 0.50 0.00 68.16
12:32:00 PM CPU %user %nice %system %iowait %idle
12:32:02 PM all 28.00 0.00 0.25 0.00 71.75
12:32:02 PM 0 33.33 0.00 0.50 0.00 66.17
12:32:02 PM 1 34.33 0.00 0.50 0.00 65.17
12:32:02 PM 2 40.00 0.00 0.00 0.00 60.00
12:32:02 PM 3 5.00 0.00 0.00 0.00 95.00
12:32:02 PM CPU %user %nice %system %iowait %idle
12:32:04 PM all 28.80 0.00 0.37 0.00 70.82
12:32:04 PM 0 30.50 0.00 0.50 0.00 69.00
12:32:04 PM 1 28.64 0.00 0.00 0.00 71.36
12:32:04 PM 2 22.89 0.00 1.00 0.00 76.12
12:32:04 PM 3 32.66 0.00 0.00 0.00 67.34
Average: CPU %user %nice %system %iowait %idle
Average: all 28.74 0.02 0.44 0.10 70.70
Average: 0 33.38 0.00 0.65 0.20 65.77
Average: 1 27.59 0.05 0.35 0.00 72.01
Average: 2 25.72 0.05 0.45 0.10 73.68
Average: 3 28.24 0.05 0.30 0.10 71.31
2.- About transactions....
420KB.seg = 3360Kb/s = 3Mb/s ( so it is not a problem of network ).
3.- The only way to increase the trhoughput is to download the PermSize in
JVM, and in this case I can download the CPU usage and after increase the
trhoughput. But, I am never able to increase more than 25% of CPU...
4.- I will check if it is a limitation of OS, I am using RHEL4.5
2009/2/25 Thomas Johnson <[email protected]>
Hmm, just ignoring jconsole for a moment, what does your OS say about
its CPU usage? Does top, prstat, acitivity monitor, or task manager
show that your Java process is also using 25% CPU? Is that 25%
combined, or is that just User space, not counting time in Kernel
space?
Once again, I'd just like to rule out a bottleneck at the network
level. Can you check how much bandwidth you're using, and compare it
to the amount of bandwidth that's available on your network? How large
are the transactions? 100tx/sec multiplied by (arbitrary) 100
kilobytes per transaction is a full 10 MBps, which will easily
saturate a 100Mbps ethernet link.
Also, does changing the thread count do anything to your Transaction
rate? Do you still only get 100 tx/sec if you increase the thread
count? What is the impact on your CPU usage when you double the thread
countL
Also, do you know if your system policy is limiting your resource
usage to a given percentage?
On Wed, Feb 25, 2009 at 11:46 AM, Toni Menendez Lopez
<[email protected]> wrote:
I am testing a Huge platform that is able to get like 5000reqxseg, and
now I
am only sending 100reqxseg, so my Load is doing nothing to the Tested
server.
About logs, I am not writing any logs in my jmeter...
May main question, and I am not able to find a solution, and I am
browsing
in google is why with jconsole the CPU of the JVM is not able to go
further
than 25% even when I have much free CPU.
Toni.
2009/2/25 Thomas Johnson <[email protected]>
Just curious -- what kind of system are you testing?
Are you doing tests against a website? You might be hitting bandwidth
limitations, meaning that the test machine's CPU has nothing to do
while it waits for the network. If you are running the server on the
same system, the server might be eating the other 75% of the CPU.
Also, if you are logging data, what kind of device are you writing
logs to? Is it a local hard drive, or a networked drive?
On Wed, Feb 25, 2009 at 11:17 AM, Toni Menendez Lopez
<[email protected]> wrote:
OK, I have a look to the links and still thinks that I don´t
underestand
quite well, I explain to you :
After lloking in the CPU´s in OS I can see that the threads are spread
in
all CPU´s look attachment ( CPU.jpg ).
After this I tried to check the JVM working mode and I go to jconsole,
and
now in this case I am not able to get more than 25% of CPU, look the
graph... (jconsole.jpg)
Do you have any idea why the JVM of the jmeter is not able to go
further
than 25% ?
Toni.
2009/2/25 <[email protected]>
Hello Toni, I think Sebb et. al. was trying to get you to understand
the
complexities of the JVM (see link). Even ANSI C daemons that clone
themselves and spawn off all new OS processes don't really get
involved
with
hardware CPUs as a plural concern.
http://www.mindprod.com/jgloss/jvm.html
Secondly, if you are concerned with JMeter and performance testing
and
the
performance of JMeter itself during serious application testing take
a
look
at this link:
http://jakarta.apache.org/jmeter/usermanual/jmeter_distributed_testing_step_by_step.pdf
Toni Menendez Lopez wrote ..
> But normal behaviour of Jmeter if I don´t change anything , does
not
permit
multi-cpu, isn´t it ?
Toni.
2009/2/24 Toni Menendez Lopez <[email protected]>
Just any clue where to find ?
Toni.
2009/2/24 sebb <[email protected]>
Check the JVM documentation?
On 24/02/2009, Toni Menendez Lopez <[email protected]> wrote:
Which parameter I can add to JVM to use more than 1 CPU ?
Toni.
2009/2/24 sebb <[email protected]>
> On 24/02/2009, Toni Menendez Lopez <[email protected]>
wrote:
> > Hello, every body....
> >
> > Just one question, Is it possible to send jmeter to work
in
multiple
> CPUs as
> > a single process, or I have to send a process for each
CPU
?
>
> JMeter is pure Java, and will use as many CPUs as the JVM
may
choose
to
> use.
>
> >
> > Toni.
> >
>
---------------------------------------------------------------------
> To unsubscribe, e-mail:
[email protected]
> For additional commands, e-mail:
[email protected]
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail:
[email protected]
For additional commands, e-mail:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
