Posted Sep 13, 2007 3:12 UTC (Thu) by JoeBuck (subscriber, #2330)
[Link]
My guess is that some of what we consider userspace, they consider part of the OS (specifically, all those cool KDE and Gnome features). I think Windows has paid a cost for the tighter integration (more security issues) but might also have seen some benefits in the power area.
Sucking power
Posted Sep 13, 2007 15:57 UTC (Thu) by rfunk (subscriber, #4054)
[Link]
Of course, in our case I would think that the Qt and Gtk+ libraries should
be able to deal with much of those problems themselves, centralizing the
problem. I suspect few people these days run apps that talk directly to
the X server.
Sucking power
Posted Sep 20, 2007 6:14 UTC (Thu) by tuna (guest, #44480)
[Link]
Those libraries are dealing with the problem, look at the new polling functions in Glib such as g_timeout_add_seconds ().
Sucking power
Posted Sep 13, 2007 16:04 UTC (Thu) by iabervon (subscriber, #722)
[Link]
My feeling is that a smaller portion of applications on Windows do anything at all when they aren't the user's main task. There's a relatively small set of applications that will interrupt you (many of which are integrated with the kernel and sleep on input from drivers instead of polling), and some that play background music, but mostly they don't do anything when they don't have focus. Linux doesn't have so much of a mechanism for telling applications whether the user cares about them (and tends to have users watching a dozen things out of the corners of their eyes), so everything has to be polling if the things it is supposed to respond to aren't things that can be blocked on.
It's also the case with Gtk+ at least that it's hard to get good results out of multiple threads (each of which blocks on a different input) all updating the display; the only easy thing that works (last time I checked) is to have a single thread poll for input having happened and take care of responding. I think Windows has a totally different locking model that requires less polling.
Of course, I run a particularly old-school UI on my laptop, and I've been getting ~14 W for interaction stuff (i.e., when I'm not actually creating significant load; how much power doing work uses is a separate issue). On my laptop, the main sources of wakeups seem to be that the trackpad's input doesn't get batched at all and is high-frequency when in use, and iwlwifi is always busy.
Sucking power
Posted Sep 16, 2007 3:36 UTC (Sun) by sobdk (guest, #38278)
[Link]
What makes you think Windows doesn't have a similar problem? With all of the crap you have to install on Windows like virus scanners, anti-spyware tools, internet security suites I'm sure the situation is much much worse. You did get me curious enough to do a little test today on my wife's machine that has both Windows Vista (pretty much a bare install with virus software) and Kubuntu.
The gist is that on the same machine Windows would idle at about 79 Watts but Kubuntu 7.04 with a kernel.org 2.6.22.6 used 69 Watts. The real kicker for me was that even though I thought we had been putting the machine to sleep in Windows all of this time it actually does nothing!
Vista ?
Posted Sep 16, 2007 9:46 UTC (Sun) by khim (subscriber, #9252)
[Link]
Why the hell you are comparing Linux with Vista ? Vista is a hog in all senses. XP or W2K are good, though (W2K or Core, but not on Pentium 4).
Vista ?
Posted Sep 16, 2007 15:05 UTC (Sun) by sobdk (guest, #38278)
[Link]
>Why the hell you are comparing Linux with Vista?
Well, it was the only Windows machine I have (Thank God!).
Honestly though do you think my comparison was unfair? I compared the latest and greatest from both the Linux world and the Windows world. Most consumers whether they like it or not will will get a new PC that has Windows Vista preinstalled. I would personally love to hear someone from Microsoft say "No wait that's not fair! Please do a comparison with our older far superior W2K and XP!"
Additionally I can tell you that at WinHEC in 2006 I sat through several fine developer (marketing?) sessions explaining how all of the suspend and resume problems of XP were going to be solved in Vista. In XP drivers (and perhaps app software) can simply veto a request for suspend. "Less asking and more Telling" was the moto for Vista as they explained that all drivers would have to get their act together and support suspend properly. Well someone didn't get the memo on my machine.