TuxOnIce in Ubuntu kernel?

AceLan Kao acelan at mail.canonical.com
Mon Oct 19 03:55:58 UTC 2009


I have applied the tuxonice patch on the karmic kernel and it works pretty
good for me.
(I had a tuxonice branch for it,
I have done a non-professional measurement of my laptop suspend/resume time.

Here is my h/w spec.
CPU: Intel Core2 Duo T5500 (1.66 GHz)
HD: Toshiba SATA2 320GB with 8MB cache

     suspend time /resume time
swsuap   56s    /    1:58s
tuxonice1 1:02s   /   50s
tuxonice2 30s   /   35s

For these three test case,
1. the swsuap case is suspended as soon as the system is booted up and it
doesn't indicate how many RAM the system consume while it suspend/resume, so
I don't have the info about that.
2. tuxonice1, the system occupied about 2.5GB RAM. I'm operating my laptop
for a while, and suddenly think I have to measure the suspend/resume time.
3. tuxonice2, the system occupied about 1.5GB RAM. Clear reboot my laptop
and open firefox with about 50 tabs and then suspend/resume.

TuxOnIce has more other features, such as suspend/resume splash screen,
reuse the suspend image, save the suspended image in a file, not in a

And there are some reasons I guess that mainstream kernel doesn't acquire
this patch is that the tuxonice patch is not only affect the kernel, there
are some config files should exist in user space and initramfs as well.

I think it would be a good feature that Ubuntu could acquire this patch and
integrate all necessary files in user space. It feels so good that
suspend/resume could be done so fast.

Anyway I could demo you what I have in the laptop at UDS, if anyone
interested in tuxonice patch.

Best regards,
AceLan Kao.

2009/10/19 Andy Whitcroft <apw at canonical.com>

> On Sun, Oct 18, 2009 at 04:59:03PM +0800, Chow Loong Jin wrote:
> > Hi Kernel Team,
> >
> > I've recently been testing out TuxOnIce on my notebook, and was rather
> surprised
> > to note the performance improvements over the kernel method/swsusp. (Less
> than
> > half a minute to a functional desktop, with swap in dm-crypt/lvm).
> >
> > On the same system, the kernel method and uswsusp both take 2-5 minutes
> to
> > resume, depending on how much memory/swap usage I had prior to
> hibernating.
> > While this might have been acceptable in the past, with Ubuntu's boot
> time
> > getting shorter and shorter (the target is 10s, IIRC), I think it's time
> our
> > hibernating/resuming time grew shorter.
> >
> > Something I did notice in my tests is that for the same system, all three
> > methods resumed from cryptsetup's password prompt to userspace in pretty
> similar
> > time frames (<30s). However, for the kernel method and uswsusp, the
> system is
> > unusable while it continues to swap out for 2-5 minutes. A user on a bug
> report
> > once did some benchmarks by running `vmstat 1` and hibernating. The
> result was
> > that the swapout rate was approximately 1MB/s after resuming. IMO this is
> rather
> > suboptimal and could be improved much.
> >
> > And hence my suggestions:
> > 1) Switch to tuxonice for hibernate/resume (tuxonice-userui can be
> improved, and
> > pm-utils already has support for tuxonice)
> > 2) Improve (u)swsusp to read out more from the swap instead of unfreezing
> tasks
> > prematurely and having the rest swapped out slowly.
> It seems that the plan is to pull the good bits of Tux-on-ice into the
> mainline kernel in an incremental manner over the next couple of releases.
> We will need to monitor this for Lucid.
> -apw
> --
> kernel-team mailing list
> kernel-team at lists.ubuntu.com
> https://lists.ubuntu.com/mailman/listinfo/kernel-team

Chia-Lin Kao(AceLan)
E-Mail: acelan.kaoATcanonical.com (s/AT/@/)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20091019/c7999d6c/attachment.html>

More information about the kernel-team mailing list