Ross Boylan wrote:
On Sun, 2007-08-19 at 01:15 +0200, Bart Samwel wrote:
Hi Ross,
This package depends on a number of hardware-specific packages
(radeontool, toshset). Shouldn't those be "recommends" or "suggests"
rather than "depends" relations?
No, not really. The trouble is that this package is installed by default
on laptops, and that it's supposed to make laptops "just work". If
anybody gets it into their heads to not install "recommends" packages,
then the package will no longer work as intended. And there's no proper
way for the package to complain.
See also: #410918, #434566.
Sorry I missed that: I was looking at acpid bug reports. Yes, this is
basically the same issue.
Does the anticipated (or has it happened?) transition of aptitude to
install recommends by default change the need to make these "depends"?
It may solve the problem for aptitude users, but then again, not
everybody uses aptitude. Also, we have an upstream package to deal with
(it's an Ubuntu package) that isn't robust against these things not
being installed. So it's going to be a lot of trouble fixing this up.
Also, it seems cleaner to me to make a meta-package that will pull in
all the dependencies. Or perhaps they could go directly in the laptop
task, since that already is a metapackage.
But not everybody actually installs the laptop task. :-) I could imagine
splitting acpi-support into acpi-support and acpi-support-laptop, where
some of the scripts would also move to acpi-support-laptop. But that
would leave a _very_ small acpi-support package. Also, again we have to
deal with the upstream, which isn't split in such a way. And as
acpi-support is really kind of a transitional package until a better
solution comes along, we're probably not going to bother creating such a
big diff with the upstream.
The acpi-support package is primarily intended for laptops. For laptops,
battery life is essential. And for power saving on laptops, setting
things like screen brightness, DPMS timeouts etc. is definitely
essential. This explains the reasons for depending on xbase-clients and
laptop-detect.
What if someone wants to use a laptop in text only mode?
Well, it works. :-) I do see that the xbase-clients dependency is a bit
broad (xbase-clients depends on ~100 packages), perhaps we can specify
the actual dependencies instead.
If you want to run acpi-support on a server system, it
will normally do no harm, but it won't do much good either.
I was afraid of that.
Well, it can do suspend/resume of course, if the hardware supports it.
Note that we haven't tested this, I don't own anything other than
laptops. :-)
Do you want
to use it to go into suspend mode or something like that? If you want to
save power on a server, you might be interested in something like
laptop-mode-tools, by itself, without acpi-support. But please,
enlighten me about your needs. I'm quite willing to think along. :-)
Thanks for the tip. In response to your later message, this message
struck me as helpful, not sarcastic.
OK, good. :-)
I'm trying to do my bit to fight global warming--without turning my
computer off.
That's one of the reasons I started working on laptop-mode-tools at some
point -- I had a server, and I wanted the disks to spin down when I
wasn't using them.
There several things I'm hoping for.
The simplest is to be able to put the computer into low-power or even
suspend mode manually. One question is how to wake it from this state.
My motherboard can do a wake triggered by the ethernet card; that way I
could recieve emails, serve web pages, etc as needed. But it would be
nice to be able to wake it in person (hit the power key? the keyboard
or mouse?).
>
The next trick would be for it to go into a low power state
automatically. I thought this "just worked," but I suspect it doesn't
work too much--I'm not even sure if my CPU and other components throttle
down when they are idle. My screen does blank, at least (using KDE).
Finally, from what I can tell some of the sleep modes essentially turn
things off. But I have stuff I like to do overnight, like run backups.
I would like it if those continued to work, either because the machine
wasn't really asleep or because it woke up at a set time. My
motherboard doesn't appear to support a timed wakeup. Maybe I could do
something via my UPS, but it's pretty dumb.
This is obviously wandering beyond a bug, but if you have any thoughts
or pointers that would be great. I've skimmed the ACPI spec, but things
are still a bit fuzzy.
I think that basically, the turning servers off and on based on demand
is a dream. It's simply never going to happen. The reason? Well, it is
pretty much impossible to detect what the "demand" on a server is. You
would have to check out what the time of the next CRON job is and set a
timed wake-up, use wake-up-on-LAN for truly on-demand stuff, you would
have to monitor network traffic and only suspend after 5 or 10 minutes
of no traffic.
A much better avenue, I think, is to use low-power always-on things.
Personally, I've solved my problem (an always-on but very-low-usage
server) by moving my server to a shared virtual server at a hosting
company. But even if you can't do that, there are two other ways of
achieving low power:
1. Switch to lower-power hardware. See for instance this setup:
http://blogs.zdnet.com/Ou/?p=275.
2. Tune your existing hardware. laptop-mode-tools is a very good
starting point for this, because it can automatically configure a
plethora of things such as screen blanking (also for terminals), hard
disk down-spinning, CPU frequency scaling etcetera, when you configure
it to do so. Not running X at all is a very good power-saver as well,
because you can keep the graphics card in a low-power terminal mode. In
addition, after you've configured your CPU frequency scaling, you should
run "powertop" to see what's keeping your CPU alive. :-)
If there are good answers to how to do power saving on a server, it
might be good for README.Debian (or the package description). But
perhaps this isn't the right package for it. I'll have a look at
laptop-mode-tools. I hadn't thought to use it, because the package
description makes it sound as if it's only for laptops.
As I said, I'm not sure how much power-saving I'm getting now, nor how
much better I can do. I hope Debian and Linux can move toward making
this automatic.
Unfortunately ACPI does not usually provide power info on servers, while
on laptops you can read out the exact power draw. It would probably be
interesting to get yourself a Kill-A-Watt ($40) or a similar power
monitoring device so that you can see exactly what your changes are
doing. Such a device is very useful for finding out what the big power
suckers in the rest of your setup / house are as well. For instance, I
found that my digital cable TV box draws 18W *in standby mode*, about
half of what my complete server used when I still had it inside the
house. :-)
Cheers,
Bart
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
