[Qemu-devel] Guest bootup time varies with simulated memory size

[Qemu-devel] Guest bootup time varies with simulated memory size

[Adnan Khaleel]

[-- Attachment #1: Type: text/plain, Size: 1012 bytes --]

In trying out various simulated memory sized in Qemu, there is a noticeable increase in guest bootup times with more memory. My metric of measuring how long linux takes to boot is probably not the most consistent but here is what I've observed.


In Qemu 0.12.5


512m ~ 4 mins
2048m ~ 7 mins


In the latest version of Qemu 0.14.0-rc0, things have improved


512m ~ 3 mins 30 secs
2048m ~ 4 mins 45 secs


As you can see, boot up time increases with more memory. My host system itself has 16G of memory so swapping is not the issue. 
Its quite likely that the guest linux kernel maybe doing something with this extra memory but I haven't looked into possible causes. I just wanted to see if the developers would know of an explanation as to whats happening as it would seem that adding more memory to the guest should improve performance. 


I will run some measurements after boot to see if there is any variation in overall simulation speed or if its just restricted to bootup.


- AK 

[-- Attachment #2: Type: text/html, Size: 1475 bytes --]

Re: [Qemu-devel] Guest bootup time varies with simulated memory size

[Mulyadi Santosa]

Hi Adnan...

On Wed, Feb 9, 2011 at 05:45, Adnan Khaleel <adnan@khaleel.us> wrote:
> In trying out various simulated memory sized in Qemu, there is a noticeable
> increase in guest bootup times with more memory. My metric of measuring how
> long linux takes to boot is probably not the most consistent but here is
> what I've observed.
> In Qemu 0.12.5
> 512m ~ 4 mins
> 2048m ~ 7 mins
> In the latest version of Qemu 0.14.0-rc0, things have improved
> 512m ~ 3 mins 30 secs
> 2048m ~ 4 mins 45 secs

Interesting number.... things I suggest for further testing:

1. what happen if you also pass "mem=<nn>" to kernel parameter?

2. if the kernel your linux distro use is enabling PAE (hence 64 GiB
support), try to fall back to non PAE one....

3. what happen if you increase the translated code cache size
(-tb-size parameter)?

anyway, it could be something related to a small patch submitted weeks
ago by somebody, which reinsert the last executed TB back to head of
the code list

-- 
regards,

Mulyadi Santosa
Freelance Linux trainer and consultant

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