Power consumption out of box

[Justin Dugger]

On 10/15/07, Alan Pope <alan at popey.com> wrote:
>
> > * Do glxgears and CPU Burn-In represent valid load tests?
>
> IMO no, not at all. A real world test might be more "realistic". But
> everyones "real world" is different. For me it might include a wifi
> access point in use with evolution open, checking mail via IMAP
> regularly. It would also include one or more ssh sessions and a browser
> or two. None of that is going to load the CPU tremendously but would use
> wifi and a little disk IO (storing mail headers for example).

I don't think people will be doing heavy file / number crunching, but
there is a slightly more valid use case: video games.  I kind of take
issue with glxgears as somehow 3d representative. As the nouveau
people pointed out, it is a benchmark, but only of buffer clearing
speed.  It doesn't put texture rendering to work, or much of the rest
of the 3d chipset.  Everyone says it's not a valid 3d benchmark, and I
suspect for power consumption purposes this is true.

> When people run systems on a battery, do they really run their systems
> flat out? Unlikely. I tend not to do video re-encoding on a system
> that's running off battery because I know it's going to eat the power.

Well, I argue that DVD playback on a widescreeen or simply a high res
laptop panel qualifies as a simple form of video processing.  DVD
playback sounds like a good idea to test, but I don't know how
testable it is.  From a power perspective, you've got the drive
itself, the software to decode it, and the (generally) crappy playback
tools that often trigger DPMS and friends.  Basic testing of usage
could be done via scripts, or worst case, VNC.  I'm not sure how
portable between releases a VNC solution would be, but I've heard
various large Operating System corporations do test in such a manner.

> That said, the test is a comparative one between releases, not a
> benchmark of real world usage, so in some ways at least it is valid.
>
> > * Should we be concerned that 5.10 represents the lowest power draw
> > from the kernel?
> >
>
> Only if the test is valid. Should we be developing a simple script that
> kicks off some standard applications (like the ones outlined above) and
> then logs battery state to a file periodically? I would recommend that
> whatever versions are tested, they should be up to date with whatever
> update-manager recommends, but other than that they should be pretty
> clean installs.

That sounds like a nifty endeavor. What do you think of the 'stress'
package?  It sounds like a nice tool for at least approximating
various workloads.  It highlights that it's not a benchmark, but I
can't see how it would hurt to use it to generate something at least
reproducible.  Even if one can't really put out something useful from
a DVD playback or gaming perspective, it'd still help to have
something _everyone_ does on battery.

> I'd like to see a comparison of the current supported releases - Dapper,
> Edgy, Feisty and (soon) Gutsy, and then further, a comparison with Gutsy
> after doing everything that powertop recommends.

Has anyone actually gone about seeing to the powertop suggestions?
Some of them aren't user friendly, but some seem fairly innocuous.
It'd be nice to document an opinion if a set of suggestions haven't
been included for sound reasons.

> > I've noticed in all their desktop screenshots, the update-manager was
> > prompting for updates. I don't know of any updates regarded to improve
> > power usage, but I think later versions will download and unpack,
> > potentially adding a long set of updates to download and unpack in the
> > background.

> You could perform the tests only after the system has settled for a
> while - after all updates have been processed, and update-manager has
> done its checks.
>
> Maybe we should have a little wiki page to gather this information?

My one qualm with putting performance data in the wiki is that tabluar
data is not very well supported.  The laptop testing project already
discovered the troubles this causes with growing users and test cases.
 As you get more reports on new laptops and new releases, managing the
wiki entry becomes painful.  Moreover, the format doesn't lend itself
well to more creative ideas like filtering for working wifi, intel
graphics, etc.  Sadly, I know of no better solution.

Justin Dugger


> Cheers,
> Al.
>
> --
> laptop-devel mailing list
> laptop-devel at lists.ubuntu.com
> https://lists.ubuntu.com/mailman/listinfo/laptop-devel