kernel 2.6.32 much slower than 2.6.31 on s2disk

kernel 2.6.32 much slower than 2.6.31 on s2disk

[Willi Mann]

Hi!

It seems to me that kernel 2.6.32 takes much longer to restore the system 
into a really responsive state, despite the fact that s2disk seems to finish 
much faster. 

I don't know to how to do a reliable benchmark on this problem, espially as 
the required time probably very much depends on the exact state of the 
frozen system. Is there any change in 2.6.32 that might cause less memory to 
be stored on the suspend device, and thus require more random disk access 
after the restore?

WM

Re: kernel 2.6.32 much slower than 2.6.31 on s2disk

[Willi Mann]

I can no longer reproduce this, there might have been some very large 
process that caused the delay. 

WM

Re: kernel 2.6.32 much slower than 2.6.31 on s2disk

[Rafael J. Wysocki]

On Wednesday 16 December 2009, Willi Mann wrote:
> Hi!
> 
> It seems to me that kernel 2.6.32 takes much longer to restore the system 
> into a really responsive state, despite the fact that s2disk seems to finish 
> much faster. 
> 
> I don't know to how to do a reliable benchmark on this problem, espially as 
> the required time probably very much depends on the exact state of the 
> frozen system. Is there any change in 2.6.32 that might cause less memory to 
> be stored on the suspend device, and thus require more random disk access 
> after the restore?

Actaully, yes, there is.

Please try to increase the value in /sys/power/image_size to approximately
1/2 of your RAM and report back (the number is in bytes).

Rafael

Re: kernel 2.6.32 much slower than 2.6.31 on s2disk

[Willi Mann]

>> I don't know to how to do a reliable benchmark on this problem, espially as 
>> the required time probably very much depends on the exact state of the 
>> frozen system. Is there any change in 2.6.32 that might cause less memory to 
>> be stored on the suspend device, and thus require more random disk access 
>> after the restore?
> 
> Actaully, yes, there is.
> 
> Please try to increase the value in /sys/power/image_size to approximately
> 1/2 of your RAM and report back (the number is in bytes).

Without changing the image_size, it especially got much better when I
downgraded QT 4.6 to QT 4.5 which does not work well with KDE 4.3 (seems
to cause memleaks).

However, image_size is already set to your recommended value (well,
approximately):

# cat /sys/power/image_size
951431086

My RAM size is 2 GB (however, I have intel graohics with shared mem, so
some part is reserved), my swap size is a little bit more than 2 GB.

I don't know what value image_size was set to when I tried first.

Note that when I reported the issue first I used Debian kernel 2.6.32-1,
(probably plain 2.6.32), while I'm now using Debian kernel 2.6.32-2
(according to the changelog 2.6.32.1)

WM

Re: kernel 2.6.32 much slower than 2.6.31 on s2disk

[Rafael J. Wysocki]

On Monday 28 December 2009, Willi Mann wrote:
> 
> >> I don't know to how to do a reliable benchmark on this problem, espially as 
> >> the required time probably very much depends on the exact state of the 
> >> frozen system. Is there any change in 2.6.32 that might cause less memory to 
> >> be stored on the suspend device, and thus require more random disk access 
> >> after the restore?
> > 
> > Actaully, yes, there is.
> > 
> > Please try to increase the value in /sys/power/image_size to approximately
> > 1/2 of your RAM and report back (the number is in bytes).
> 
> Without changing the image_size, it especially got much better when I
> downgraded QT 4.6 to QT 4.5 which does not work well with KDE 4.3 (seems
> to cause memleaks).
> 
> However, image_size is already set to your recommended value (well,
> approximately):
> 
> # cat /sys/power/image_size
> 951431086
> 
> My RAM size is 2 GB (however, I have intel graohics with shared mem, so
> some part is reserved), my swap size is a little bit more than 2 GB.
> 
> I don't know what value image_size was set to when I tried first.
> 
> Note that when I reported the issue first I used Debian kernel 2.6.32-1,
> (probably plain 2.6.32), while I'm now using Debian kernel 2.6.32-2
> (according to the changelog 2.6.32.1)

So, is it still a problem for you?

Rafael

Re: kernel 2.6.32 much slower than 2.6.31 on s2disk

[Willi Mann]

>> Note that when I reported the issue first I used Debian kernel 2.6.32-1,
>> (probably plain 2.6.32), while I'm now using Debian kernel 2.6.32-2
>> (according to the changelog 2.6.32.1)
> 
> So, is it still a problem for you?

No, but I'm not sure why it is no longer a problem (kernel or userland
change?).

WM

Re: kernel 2.6.32 much slower than 2.6.31 on s2disk

[Rafael J. Wysocki]

On Thursday 31 December 2009, Willi Mann wrote:
> 
> >> Note that when I reported the issue first I used Debian kernel 2.6.32-1,
> >> (probably plain 2.6.32), while I'm now using Debian kernel 2.6.32-2
> >> (according to the changelog 2.6.32.1)
> > 
> > So, is it still a problem for you?
> 
> No, but I'm not sure why it is no longer a problem (kernel or userland
> change?).

Hard to say for sure, but since s2disk hasn't changed for a long time, I guess
it's a result of the kernel change.

Rafael