[Qemu-devel] Irregular response times

[Qemu-devel] Irregular response times

[Pascal J. Bourguignon]

I've got various Qemu running linux (gentoo or debian, all running the
same kernel).  On one of them,  I get very irregular response time.

During a few seconds, it works normally.  Eg. thru a ssh connection, I
can type commands at bash, and I get the character echoed right away.
And then, for a few tens of seconds, it hangs, with no output and no
way to interrupt it or whatever.   The qemu process is in S state,
waiting in poll_s WCHAN.

The same occurs on the console too.


Any hints?
-- 
__Pascal Bourguignon__                     http://www.informatimago.com/

[Qemu-devel] Re: Irregular response times

[Pascal J. Bourguignon]

pjb@informatimago.com (Pascal J. Bourguignon) writes:

> I've got various Qemu running linux (gentoo or debian, all running the
> same kernel).  On one of them,  I get very irregular response time.
>
> During a few seconds, it works normally.  Eg. thru a ssh connection, I
> can type commands at bash, and I get the character echoed right away.
> And then, for a few tens of seconds, it hangs, with no output and no
> way to interrupt it or whatever.   The qemu process is in S state,
> waiting in poll_s WCHAN.
>
> The same occurs on the console too.

Eventually, I could run emacs thru ssh, and it looks like these
response time irregularities don't occur within emacs. 


> Any hints?

-- 
__Pascal Bourguignon__                     http://www.informatimago.com/

Re: [Qemu-devel] Re: Irregular response times

[Mulyadi Santosa]

On Sun, Jul 4, 2010 at 06:42, Pascal J. Bourguignon
<pjb@informatimago.com> wrote:
> pjb@informatimago.com (Pascal J. Bourguignon) writes:
>
>> I've got various Qemu running linux (gentoo or debian, all running the
>> same kernel).  On one of them,  I get very irregular response time.
>>
>> During a few seconds, it works normally.  Eg. thru a ssh connection, I
>> can type commands at bash, and I get the character echoed right away.
>> And then, for a few tens of seconds, it hangs, with no output and no
>> way to interrupt it or whatever.   The qemu process is in S state,
>> waiting in poll_s WCHAN.

Did you notice the qemu or kvm version? Also, is it happen in pure
qemu or qemu utilizing kvm? or both?

I somewhat "smell" buggy spurious interrupt handling..... try to set
your processor to use constant clock frequency.

And....few month ago, I learned quite funny tricks. Sometimes qemu/kvm
window hangs if I place it on different workspace other than my
current workspace in my GNOME desktop. Moving out focus from Qemu's
workspace somehow make it hang for few seconds (qemu-kvm 0.10.6). So,
the trick is, always place it on your current active workspace.

-- 
regards,

Mulyadi Santosa
Freelance Linux trainer and consultant

blog: the-hydra.blogspot.com
training: mulyaditraining.blogspot.com

Re: [Qemu-devel] Re: Irregular response times

[Pascal J. Bourguignon]

Mulyadi Santosa <mulyadi.santosa@gmail.com> writes:

> On Sun, Jul 4, 2010 at 06:42, Pascal J. Bourguignon
> <pjb@informatimago.com> wrote:
>> pjb@informatimago.com (Pascal J. Bourguignon) writes:
>>
>>> I've got various Qemu running linux (gentoo or debian, all running the
>>> same kernel).  On one of them,  I get very irregular response time.
>>>
>>> During a few seconds, it works normally.  Eg. thru a ssh connection, I
>>> can type commands at bash, and I get the character echoed right away.
>>> And then, for a few tens of seconds, it hangs, with no output and no
>>> way to interrupt it or whatever.   The qemu process is in S state,
>>> waiting in poll_s WCHAN.
>
> Did you notice the qemu or kvm version? Also, is it happen in pure
> qemu or qemu utilizing kvm? or both?

This is 

% qemu --version
QEMU PC emulator version 0.12.4 (qemu-kvm-0.12.4), Copyright (c) 2003-2008 Fabrice Bellard

% uname -a
Linux kuiper 2.6.34-gentoo-c5-kvm #8 SMP Fri Jul 2 19:19:27 CEST 2010 x86_64 Intel(R) Core(TM) i7 CPU 950 @ 3.07GHz GenuineIntel GNU/Linux

% ls -l /dev/kvm
crw-rw---- 1 root kvm 10, 232 Jul  2 19:28 /dev/kvm

% cat /domU/bin/qemu-hubble
#!/bin/bash
vm=hubble
nic_model=e1000

cpu=coreduo
ncpus=2
ncores=2
mem=2G
# video=(-vga std)
# video=(-nographic)
video=(-curses)

if [ -r /dev/kvm ] ; then
    kvm='-enable-kvm'
else
    kvm=''
fi

exec qemu $kvm \
    -cpu ${cpu}   -smp ${ncpus},cores=${ncores}  -m ${mem}  ${video[@]} \
    -boot menu=on \
    -net nic,model=${nic_model} -net tap \
    -hda /domU/images/${vm}.boot.disk \
    -hdb /domU/images/${vm}.ext3 \
    -hdc /domU/images/${vm}.swap \



> I somewhat "smell" buggy spurious interrupt handling..... try to set
> your processor to use constant clock frequency.

I think it is:

$ zcat /proc/config.gz |grep HZ
# CONFIG_NO_HZ is not set
# CONFIG_HZ_100 is not set
# CONFIG_HZ_250 is not set
# CONFIG_HZ_300 is not set
CONFIG_HZ_1000=y
CONFIG_HZ=1000
# CONFIG_MACHZ_WDT is not set



That said, I do also have problems with some conftest (eg.
stackoverflow_handler / setcontext or stackoverflow_handler /
sigaltstack)  that hang on this x86_64 host.  Mostly when I boot a Xen
kernel, but I think I saw it once also on this Linux+KVM kernel
(without Xen).  So indeed, it sounds like some problem with handling
signals.



> And....few month ago, I learned quite funny tricks. Sometimes qemu/kvm
> window hangs if I place it on different workspace other than my
> current workspace in my GNOME desktop. Moving out focus from Qemu's
> workspace somehow make it hang for few seconds (qemu-kvm 0.10.6). So,
> the trick is, always place it on your current active workspace.

Well, I use ratpoison on X, and I launch the Qemu VM from screen in an
xterm.  These machines are launched with -curses.


-- 
__Pascal Bourguignon__                     http://www.informatimago.com/

[Qemu-devel] Re: Irregular response times

[Pascal J. Bourguignon]

pjb@informatimago.com (Pascal J. Bourguignon) writes:

> Mulyadi Santosa <mulyadi.santosa@gmail.com> writes:
>
>> On Sun, Jul 4, 2010 at 06:42, Pascal J. Bourguignon
>> <pjb@informatimago.com> wrote:
>>> pjb@informatimago.com (Pascal J. Bourguignon) writes:
>>>
>>>> I've got various Qemu running linux (gentoo or debian, all running the
>>>> same kernel).  On one of them,  I get very irregular response time.
>>>>
>>>> During a few seconds, it works normally.  Eg. thru a ssh connection, I
>>>> can type commands at bash, and I get the character echoed right away.
>>>> And then, for a few tens of seconds, it hangs, with no output and no
>>>> way to interrupt it or whatever.   The qemu process is in S state,
>>>> waiting in poll_s WCHAN.
> exec qemu $kvm \
>     -cpu ${cpu}   -smp ${ncpus},cores=${ncores}  -m ${mem}  ${video[@]} \
>     -boot menu=on \
>     -net nic,model=${nic_model} -net tap \
>     -hda /domU/images/${vm}.boot.disk \
>     -hdb /domU/images/${vm}.ext3 \
>     -hdc /domU/images/${vm}.swap \


I eventually found the problem:  all qemu instances will use the same
MAC address for their NIC.   When I set different MACs to the
different VMs, everything works well: 

mac=52:54:0:0:0:1 # or 2, 3, ...
...
    -net nic,macaddr=${mac},model=${nic_model} -net tap \


-- 
__Pascal Bourguignon__                     http://www.informatimago.com/