Not too long ago I purchased a P3 International Kill-a-Watt Electricity Usage Monitor. One of the reasons I purchased it was so I could see how much power my computers were using especially since we tend to have 2-3 computers running at all times. Now as you may know I'm a huge Linux fan so you may be surprised to learn that one of our five computers actually dual boots Windows Vista and Ubuntu Linux (my wife Katina says she needs Windows for some of her software for work). This seemed like the perfect opportunity to do a Windows Vista vs Linux power comparison!
Before I get into the number let me provide a little history in power management in Linux. For the longest time Linux has found its home in the server market where power management hasn't been a top priority. When I first got my IBM ThinkPad T20 notebook I found that I could get about four hours of battery life out of Windows XP and only two hours from Linux. Suspend to RAM and suspend to disk have always been (and still are) a bit shaky. If you are lucky and had well supported hardware like my T20 then suspend just works, but otherwise (like the rest of my machines) you are out of luck. Recently though things in the Linux world are starting to change. While suspend still generally sucks in Linux the fine folks at Intel have created PowerTOP an application to show you what programs and drivers are using the most power on your system. This combined with the dyntick features that recently went into the Linux kernel are allowing Linux to start making some major power improvements.
Now even though Katina probably spends about 50/50 of her time spent between Linux and Windows I thought I would first check how much power her machine uses when it is in Linux. For the record her machine is a custom built Core 2 Duo E6300 with 1 GB of RAM desktop with a crappy ECS motherboard. I plugged her machine into my Kill-a-Watt power meter and booted into Kubuntu Linux 7.04. The only customization that I have made to her machine was to upgrade her machine to the 2.6.22.6 kernel with dynticks enabled. As previously mentioned suspend to RAM does not work on this machine. After booting the machine I let things settle for 15-20 minutes then checked the power meter to see how many watts it was using. Katina surfed the web for a while and I periodically checked the power usage from time to time conclusion:
| Idle (Observed W) | Average (W) | Total Power (KWH) | Time Tested(HH:MM) |
|---|---|---|---|
| 69 | 70.1 | 0.23 | 3:17 |
| Idle (Observed W) | Average (W) | Total Power (KWH) | Time Tested(HH:MM) |
|---|---|---|---|
| 79 | 79.9 | 0.61 | 7:38 |
The great news from my point of view is that things can only get better from here in the Linux world. Linux really has a much greater advantage when it comes to improving power management, since we have the source code for everything. Using tools like powertop we can not only find who is using all of the power but anyone can fix the problem. In the Windows world we can only hope that everyone who writes software cares enough to consider power management (Ha).
Comments
delurking day
happy de-lurking day...when you gonna get around to testing out these lovely suggestions, babes?
These results are quite
These results are quite interesting, but I noticed one problem.
Your running times for the Linux and Windows power consumption were wildly different. Three hours to seven? It's possible (though unlikely) that this may have somehow skewed the results. Running the test again for only 3 hours for both would be better.
And, ideally, running the test for Linux, Windows Vista, and Windows XP would also be nice. Just to see how badly Vista has screwed the pooch.
An interesting report. It's
An interesting report. It's good to see that Linux is getting better (or maybe it's just that Windows is getting worse!). I just have one nitpick:
In your tables, KWH (should be 'kWh') is a measure of energy consumed, not power. Power is measured in watts (E = P x t).
Vista? Try XP
As khim requests, could you do a similar comparison with winXP or even w2k? I know that installing the buggers is a pain, but the results could actually be more interesting for a tried and tested Microsoft OS.
Besides, your wife might be happier with it. And also your power usage might improve.
There's the problem of getting it. You have probably some licenses lying around for some of your servers. I would like to do the test myself but sadly I don't have a usage monitor.
XP or 2K
Well, I'm sure I have the Windows licenses, but I'm a bit too lazy to reinstall Windows. I'll see if I can find another machine and run some more tests and post the results perhaps later today or tomorrow.