[Qemu-devel] Problem with high load average when using -nographic with qemu in background

[Qemu-devel] Problem with high load average when using -nographic with qemu in background

[Phil Rasch]

Sorry to bother you with this. I did try the qemu forums first.

I am having trouble running a trixbox/centos guest under an ubuntu host. Both 
completely uptodate installs of the OS's. I am using Qemu 0.8.2 with 
kqemu-1.3.0pre9.

If I start the qemu without -nographic interactively the VM runs well, and the 
hostOS maintains a load average of about 0.1

Trouble comes when I add -nographic to the command line, and redirect stdout 
and stderrout to a logfile. (I do this so I can start qemu from during the 
boot up process and run the VM in background all the time.

If I start qemu like this (from an interactive shell (tcsh))
qemu -m 256 -localtime -net nic -net tap -nographic trixbox.img \ 
>&! /tmp/qemulog &

It will stop on
 + Suspended[tty output]

If I also redirect stdin (eg adding "< /dev/null") it then starts and runs 
fine, but the hostOS load climbs to 1.0 and higher (depending on what the 
guest is doing, and the performance of the VM is seriously impaired.

I can ssh into the machine under both scenarios and see that the guest thinks 
it is fine, but it is really burdening the hostOS.

Looking through /var/log/messages on the guest shows that I have (I think) 
successfully set io to go through /dev/ttyS0, eg.

Jan  6 09:46:49 asterisk1 kernel: Kernel command line: ro console=ttyS0,9600n8 
vga=normal root=LABEL=/

Can anyone offer advice?

Phil

Re: [Qemu-devel] Problem with high load average when using -nographic with qemu in background

[Mulyadi Santosa]

Hi Phil
> If I start qemu like this (from an interactive shell (tcsh))
> qemu -m 256 -localtime -net nic -net tap -nographic trixbox.img \ 
>   
>> &! /tmp/qemulog &
>>     
I never did this before, but maybe you need to take a look on -serial 
option. Maybe you can use it to redirect the serial output onto FIFO and 
then read the other end point of this FIFO into a file

About the high load, I guess that happens when login prompt appears. It 
keeps polling for input but never get it and feeding /dev/null just made 
it "eating" endless null.

I hope this hint is useful.

regards,

Mulyadi

Re: [Qemu-devel] Problem with high load average when using -nographic with qemu in background

[Paul Brook]

On Saturday 06 January 2007 17:08, Phil Rasch wrote:
> Sorry to bother you with this. I did try the qemu forums first.
>
> I am having trouble running a trixbox/centos guest under an ubuntu host.
> Both completely uptodate installs of the OS's. I am using Qemu 0.8.2 with
> kqemu-1.3.0pre9.

1) Try disabling kqemu. We can't help you with that.
2) I'm fairly sure I already fixed this in cvs.

Paul

Re: [Qemu-devel] Problem with high load average when using -nographic with qemu in background

[Phil Rasch]

Now that I know that I didn't completely miss the point, and may even
have tried a reasonable thing, I have a question and a workaround.

1) Question: How do other people do this sort of thing? That is, start a
virtual machine when the hostmachine is booted automatically? Or dont
they worry about the load/performance issues?

2) Workaround: I now start qemu up inside a screen session at boot. This
seems to work fine, the load stays low, and performance is OK. I dont
need to depend upon X11 being up. I can connect from a regular console,
or from an xterm.

Thanks again

Phil

 

Re: [Qemu-devel] Problem with high load average when using -nographic with qemu in background

[Paul Brook]

On Sunday 07 January 2007 05:03, Phil Rasch wrote:
> Now that I know that I didn't completely miss the point, and may even
> have tried a reasonable thing, I have a question and a workaround.
>
> 1) Question: How do other people do this sort of thing? That is, start a
> virtual machine when the hostmachine is booted automatically? Or dont
> they worry about the load/performance issues?

I fixed qemu so it wasn't broken.

Paul