S3 Standby State Done Right

For Earth Day, Cameron Butterfield has written in with a pointer to his article on how to get your Windows PC into S3 sleep, and why you want to. It covers the question of how to take advantage of this extremely low-power mode even when your machine is an "always on" file server, remote desktop, or VNC server.

Also available on Linux

[Anonymous Coward]

It doesn't seem to be a hot topic because I couldn't google a definitive page. There were lots of pages for individual computers or distros though.

The documentation is probably on your own computer at: /usr/src/linux/Documentation/power/ ... The exact file on my system is states.txt but it also seems to exist on other distros as suspend.txt

Re:And Linux?

[BACbKA]

Gentoo's Power Management Guide [gentoo.org] is a bit gentoo-centric, but most things carry to another distribution easily.

dumppo.exe the Microsoft Power Tool

[MSRedfox]

Windows XP will often times not give s3 suspend as an option even when turned on in BIOS. But with Microsofts dumppo.exe utility you can force it to use an S3 or S4 state. ftp://ftp.microsoft.com/products/Oemtest/v1.1/WOST est/Tools/Acpi/dumppo.exe [microsoft.com] To force it to S3, run this under command prompt "dumppo admin minsleep=s3"

Re:network broadcast traffic

[Bozzio]

yes it will.
Or it does for me. Even if the computer is alone on the router. It seems my router occasionally broadcasts something and wakes up all my computers.

I've switched to using the magic packet alternative. The only problem is that since my server PC is behind my router, I have to SSH into the router and sent the magic packet from there. ICKY.

I hear other routers (mine is a Linksys WRT54GS) will let you WOL remotely. Normally, you just send your magic packet to the router and set up the router to convert it to a broadcast.

If I remember correctly, a magic packet is just a packet with the correct header and the client's MAC address broadcast to the network.

Re:Laptops?

[Southpaw018]

Most laptops come preconfigured to take advantage of most of the stuff in the article, though it wouldn't hurt to check. The last few new Dell laptops we've purchased at my organization default to S1 after a few minutes, S3 if you close the lid or hit sleep, and S4/S5 for shutdown.

Re:S3 and MCE

[MSRedfox]

Under MCE, I use the MCE Standy Tool. MCE has a bad habit of waking up to record a show and then not returning to standby afterwards. This can result in the computer staying on all day instead of just 1/2 hour to record a small show. The Standby Tool has features to help MCE handle power management in better ways then Windows default methods. It makes me wonder why Microsoft couldn't get things to work as smoothly as this 3rd party software. http://www.xs4all.nl/~hveijk/mst/indexe.htm [xs4all.nl]

Bad Assumption

[Anonymous Coward]

* I calculated (24 hours per day) * 30 days a week [sic] = 720 hours
* Power bills are generally measured in kilowatt hours or "kW/h"s. Power rates might be as much as $0.12 per kW/h
* Our total cost of having the computer on 24/7 for the month in this scenario would be as follows:
* .4 kW (400watts) * 720 Hours * $0.12kW/h = $34.56

The 400 watts per hour is a really poor assumption. First, the average home PC wouldn't use 400 watts at peak, let alone continuously. Secondly, the 400 watts would almost never be continuous. Even a PC left on overnight will use far less power then one being actively used during the day. My power costs almost $0.11 per kilowatt-hour, and I have a power bill during non-summer months (damn AC) of about $45-$55 dollars with two desktop computers running 24/7. I honestly wouldn't be surprised if most of this came from my computers, but it certainly isn't as much as the article makes it sound.

That said, it is a good article on how to keep the "instant-on" without using excess power.

Re:CPUIDLE

[ettlz]

Back in the pre-NT-based days, perhaps. Most modern operating system kernels issue the HLT instruction plus some extra power management jiggery-pokery to the CPU when it's not being used at max anyway. (Check out /proc/acpi/processor/CPU*/power under Linux.)

Re:FreeBSD

[evilviper]

S3 mode is entered by running "acpiconf -s 3"

All available options can be listed by running "sysctl -a hw.acpi" and included in /etc/sysctl.conf to be automatically set upon boot-up. Basically you'll only need "hw.acpi.reset_video=" set to 0 or 1 depending on your system.

If you need to unload modules or any other action before suspending, see /etc/rc.suspend. Put the opposite commands in /etc/rc.resume.

That should be everything you need. Either your hardware will work, or it won't. In the latter case, strip your system down to nothing but video, and try different video cards. Then add a piece at a time to see what's causing problems.

Re:What about wireless ?

[Guy Harris]

There was some useful info in this article about configuring your network adapter to support wake-on-lan, but what about wireless adapters? In my experience they don't seem to support WOL or any equivalent.

At least at one point, I found one 802.11 adapter or chipset that supported OnNow-style [microsoft.com] wakeup, but I don't know whether drivers supported that.

You'd have to keep the radio on, though, which means there's some power you can't save.

Is there any sort of WOL capabilities in the new 802.11n?

That's probably more of a chipset issue than a protocol/PHY issue, so I'm not sure there'd be any chanages in 802.11n - unless there's some radio-layer changes to allow the receiver to run in a low-power mode capable of receiving a wakeup indication.

Re:S3 and MCE

[MSRedfox]

It happened after watching live TV for me. If I was watching a recording or DVD and suspended, it would re-suspend after recording new shows just fine. But if I pressed the PC suspend button on my remote after watching live TV, the next time it woke up to record a show, the PC would stay on after the recording was done. Even if I pressed the pause or stop button before suspending it would still not power back off correctly. For a while there, I would always start playing a recording before suspending the PC. The Standy Tool fixed the issue and I haven't had any weird suspend problems since then.

Re:Calculations are a bit off

[pimterry]

I have a power monitor thing on the socket for my home server (it's just a box, no screen keyboard etc) and right now it's using 132 Watts downloading torrents and web serving (mostly as a web dev test site, so probably not really doing any work). It's a 3Ghz P4 too, so it's probably not as power efficient as it could be.

400 watts has got to be way off.

Re:And Linux?

[rduke15]

See this article: Debian HOW-TO : CPU power management [technowizah.com]. I used the info to configure a couple of Poweredge 860 server. Most of the time, it's at a CPU speed of 300Mhz instead of 3 Ghz. That saves quite some power, and you cannot notice the difference in speed.

Re:dumppo.exe the Microsoft Power Tool

[ergo98]

I wrote about the power consumption of S1 versus S3 sleep [yafla.com], and as you mentioned dumppo.exe was the enabling-tool that let me take advantage of this great bit of functionality.

The biggest downside of S3 sleep is that about 1 out of every 200 recovers or thereabouts it completely fails to come back, thought that's probably a mainboard issue more than an OS or technology issue.

Oh, and a great little helper app if you use S3 is WakeUpOnStandBy [dennisbabkin.com]. It allows you to configure a machine to "come alive" at scheduled times, even from an S3 sleep (apparently the BIOS supports configured wake-up times, and this app knows to tell it to wake up accordingly just as going to sleep). Very helpful little app -- I have my PC set to come alive in the morning when I know I'll be remoting in.

Oh, and rather than waking up on all network traffic, as the article recommends, it's far better to wake up on WakeOnLan packets. There are lots of resources out there for that.

Re:want a "file transfer" powerdown mode.

[ivan256]

You need to write to the disk every few seconds, to maintain a consistent state, with or without a journaling file system.

No you don't, and in fact if you mount your filesystem read-only, or noatime, and run noflushd your hard drives can spin down indefinitely as long as your dataset fits in memory. I used to get 8-9 hours out of the battery on my PowerBook G3 using this method and low screen brightness.

Of course, if you are writing to files and you do this and then lose power, you lose data... But you could store the files you are working on in flash to avoid this.

Re: S3 Standby State Done Right

[bi_boy]

Slightly off tangent, but hibernation (S4) fails in WinXP SP2 if you have more than 1GB of RAM. [microsoft.com]

Works just fine for me. Probably because I installed the udpate mentioned in the resolution section of the article sometime last year.

Re:Calculations are a bit off

[Anonymous Coward]

average P4 class Desktops consume about 100W in idle
average centrino laptops consume about 25W in idle.

Re:want a "file transfer" powerdown mode.

[Graff]

plasmacutter had this to say:

Ive noticed all companies, including apple, whose products i use, are giving you only a black and white choice. you either have the computer awake or its fully asleep.

i'd like to have a low power transfer mode, where the cpu is reduced (to 1 core at say 500 mhz), the monitor is turned off, and as much memory as possible is dedicated to the apps which are doing intensive file reads/writes. this will allow the hard drives to be used less by caching the files in ram and pulsing the hard disk.

You mean something like this?

http://docs.info.apple.com/jarticle.html?path=Mac/ 10.4/en/mh1669.html [apple.com]

Just a snippet from that page:

Some models support the Automatic setting, which allows your computer to switch rapidly back and forth between the Highest and Reduced settings to optimize energy use, depending on how much work the processor is doing.

This is basically what you are asking for. Your computer will automatically scale the processors according to the tasks you have running. I believe just about all modern Macintoshes support the Automatic setting.

400 watts is high

[jbengt]

I work as an HVAC engineer, and I have to take into account the PCs when designing air conditioning for an office. I figure 200 to 250 watts per workstation; that is supposed to take into account average usage including everything: the PC, monitor, peripherals, task lighting, occassional printers, etc. I've been told that this is too high, but my career has spanned a lot of changes - dummy terminals, energy inefficient monitors, heavy duty PC workstations, efficient but larger monitors, LCD monitors, thin clients, etc. - so I tend to take the conservative approach and assume that it can change again to higher wattages within the lifetime of the AC system. Power consumption of devices keeps on being improved, but instead of using less power, PCs do more with the same amount of power. If your PC has a 500 watt power supply it would probably never use much more than 400 watts (you need some safety margin) and it would probably use, on average, less than half that while working hard. With modern PCs it could easily use less than 50 watts when idle.

Re:Calculations are a bit off

[Michael Woodhams]

Those 600+ W power supplies are purely for people with inferiority complexes about other aspects of their lives/bodies. Here's [silentpcreview.com] a discussion about how much you can run on a 300W PSU. 300W suffices for a modern high-end CPU plus high-end GPU plus a bunch of drives, when under heavy load. Even a high end system will idle at around 150W. A more sensible system is probably idling around 80W.

NOTE: All the figures above are *not* including losses in the PSU. A modern PSU should be about 7 5% efficient, so increase the above by 1/3 to make them comparable to the 400W number in the article.

kWh

[Anonymous Coward]

Energy isn't measured in kW/h. I'm not sure what a kW/h would meassure. The article writer made this mistake and now everyone on /. is doing the same thing.

Energy is measured in kWh (or kilowatt-hours), which is one kilowatt of power used for the length of one hour.

Nice FUD

[The Bungi]

Anything M$ touches is shit

Oh yeah.

Bill Gate's memo

That's an interesting email from 1999. Myself, I've been known to send emails to the tone of "how can we prevent the competition from leeching on our multi-million dollar R&D investment with our technology partners", but OK.

Would you like to point me to the follow up email from Eric Rudder that says "Hi Bill - As you requested, we've made the ACPI extensions specific to Windows so no one else can implement them. Cheers!" I can't seem to find it.

Oh, wait - here's ACPIfor Linux [sourceforge.net] and ACPI for FreeBSD [freebsd.org]. Indeed, here's a quote from the WP entry:

The Advanced Configuration and Power Interface (ACPI) specification is an open industry standard first released in December 1996 developed by HP, Intel, Microsoft, Phoenix and Toshiba that defines common interfaces for hardware recognition, motherboard and device configuration and power management.

Now, ACPI has its shortcomings. It's complicated. It might not be your ideal of a standard. But it is an open standard, which Linux indeed implements. It might be broken in some ways in Linux as it is in Windows, but implemented it is. It's an important standard because it takes hardware out of the equation, which is important for a general OS that's supposed to support a wide range of it.

I still use APM for the most part

Really? That's also a Microsoft-defined standard [wikipedia.org] (along with Intel):

Advanced Power Management (APM) is an API developed by Intel and Microsoft

Is that standard "shit" as well? And if you all these standards from Microsoft are "shit", then why do you use them at all? You use Linux, right? Why don't you come up with your own standard and give it to the free software world so they can stop using all these "shit" open standards that Microsoft has bothered to make open for anyone to use? Which reminds me, I'd love to see that other email about ACPI I mentioned. Thanks.

To see available Sleep modes from Windows Prompt

[icepick72]

From command window:

powercfg -a
Works for both XP and Vista. Tells you what's available and what's not (S1, S2, S3,...) Vista tells you why something isn't support.

Got info from this page [tech-recipes.com]

Re:And Linux?

[smorar]

I found an awesome blog post the other day which explained how to get a desktop linux box into S3 suspend. It's all about your BIOS settings. Now, I can turn my pc 'off', and it takes less than 5s to resume. http://shallowsky.com/blog/linux/desktop-suspend.h tml [shallowsky.com]

Re:Bad Assumption

[TheThiefMaster]

It depends on the power supply. The cheap ones can be as bad as 50% efficient, in which case a 200W pc would draw 400W from the wall.

Incidentally I have two PCs (with >85% efficiency PSUs) and a 19" CRT monitor plugged through a power-meter right now and they are drawing 510W total, and 425W if I turn off the monitor.

One of the PCs is a dual socket A machine with cpus that won't go below 60C despite some really powerful air cooling, and the other is an AMD A64-X2 3800+ with ATI X1900XTX. Both are fairly beefy by modern standards, and from previous tests they draw about the same power (so about 210W each). If I had cheap power-supplies in these PCs (eg qTec Gold-painted triple-fan 650W) then I would be using their estimated 400W per pc, but as I went for quality (in this case Tagan 480W) I can run two pcs for that.

Thankfully I only run one of these PCs 24/7, so it costs me about $210 per year for that one, and about 1/3 of that for the other.

If you have a crap power-supply now, then you can save the cost of a new, good one in about 6 months in the electricity you use (if you have your pc on 24/7).

Re:Nice FUD

[xenocide2]

Unfortunately, the world is far more complicated than you'd like. Matthew Garrett has [livejournal.com]extensively [advogato.org] about the subject. Truth is, ACPI a standard that nobody follows intelligently. Garrett writes about how part of the spec involves an interpreted machine code called DSDT (this already sounds like a recipe for disaster) that is used to guide actions. The problem is two fold:

• DSDT's are buggy (go figure)
  
• The common method for fixing a broken DSDT is to patch it after the machine has booted via some driver
  

Microsoft has troubles with vendors who don't care much about suspend resume functionality.

The solution isn't to go out and make yet another spec that vendors wont follow intelligently. The solution is vertical integration. Apple does it, and they can know everything they want to about their hardware. And open source software like Linux also offers the potential to do so. Dell potentially has the tools to make their Linux offering compete. I've been hoping one of the Linux laptop vendors springing up would move towards speccing their own laptops but it hasn't quite happened yet (that I know of).

Re:network broadcast traffic

[SplatMan_DK]

I do not have the same problem, and I am allready using S3 powermode.

A "broadcast" package is not a package with a single specific receiver, so why would a machine in S3 mode wake up when it detects a broadcast package? The whole point is to make sure the machine only wakes up from LAN access when there is traffic directed specifically for that interface/address?

Re:And Linux?

[Thomas Shaddack]

A way to make a computer quieter is replacing the smaller fans with somewhat bigger ones, and slow down their rotation (eg. with a suitable series resistor). The aim is to get comparable airflow over the heatsink with lower fan blade speed, which means less turbulence over them, which means much less noise.

Re:Nice FUD

[xenocide2]

That post looks a bit screwed up that I made, but he's a central member of Ubuntu, a member of the laptop, kernel, and acpi teams and one of 4 members of the technical board. I had hoped his insightful analysis would have been enough, but it seems I botched a link or two. To make up for it, here's a video [linux.org.au] of him detailing how hacking acpi is done.

Re:APM Sucks too.

[The Bungi]

Oh twitter, you're going to love this. Here's an article [advogato.org] by the guy that actually writes this stuff for your operating system. I'd like you to go through that article and please share with us where the guy that actually writes the stuff blames Bill Gates or "M$" for how ACPI works (or not). As opposed to just the general sense of "this stuff is hard" I get from it.

Once you're done getting an education, I'd like for you to explain how "M$" allegedly sabotaged ACPI on Linux. You pointed [slashdot.org] to an eight-year old email from Bill Gates that, if anything, proves Microsoft did not do anything to impact the implementation of ACPI in Linux. Seriously, just in case your FSF distortion field is turned up too high, that's exactly what you are proving by linking to that email. You have ACPI in Linux. It might be as broken as it is on Windows, but you have it. You realize that, yes? God, please tell me you realize that?