Do Any Companies Power Down at Night?

An anonymous reader writes "My Health Sciences Campus has about 8,000 desktop computers, and on any given night about half of them are left on. I know this because I track all the MAC addresses in case there is a virus outbreak. Aside from the current fad of 'being green', has anyone had any success in encouraging users to power-down at night? You could potentially eliminate running bots, protect yourself from the next virus outbreak, keep your data safe, etc. Do security concerns and power consumption issues matter enough to do this?"

Aside from being green...

[flatulus]

I won't go into the green topic. But here's a suggestion: Why don't you just shut down ethernet switches and routers at night? That would be just as effective at halting propagation of virii/bots, and would be much easier to effect.

And improved employee morale could result as well, since what would be the point of working late? :-)

Re:Aside from being green...

[Bwian_of_Nazareth]

The answer is simple. Tiny minority of the computers that are on could still be used by someone doing something important. You do not want to cut them off from the network.

Re:Aside from being green...

[Kjella]

Amen to that. I doubt anyone really wants to pull an all-nighter, they're there because there's a deadline coming or a problem that must be resolved right now. And whether true or false, the first time anyone uses the excuse "Well, I was ready to pull an all-nighter but from 10pm the network was down" the IT department will have their ass chewed out. PCs inactive -> PCs hibernate is ok, but even then you need a simple way to disable it. On several occasions I've visited copmanies that had boxes which were "don't touch - accessed remotely by VPN" where the user couldn't just unhibernate it in the morning. Plus funny stuff like updates, virus scans and backup (if applicable) that probably runs at some ungodly hour which means you need to wake them first or lose most of the downtime, run those and put them back into sleep. Sure, maybe you could get every PC do to this reliably but I think the administration and scripting of that will cost you quite a bit.

Re:

[totally bogus dude]

Wake on LAN requires a magic packet, so browse list refreshes shouldn't be a problem. A PC directly connected to the internet might be wakeable if people send it a WoL packet, and I guess that's a fairly intelligent thing for a network scanner to do if you're looking to infect as many PCs as possible, but would probably be difficult.

From everyone's favourite almost-an-encyclopedia [wikipedia.org]:

The Magic Packet is a broadcast frame, transmitted over port 0 (Historically the most common port used), or 7 or 9 (becoming the most common ports used). It can be sent over a variety of connectionless protocols (UDP, IPX) but UDP is most commonly used. The data that is contained in a Magic Packet is the defined constant as represented in hexadecimal: FF FF FF FF FF FF followed by sixteen repetitions of the target computer's MAC address, possibly followed by a four or six byte password.

It's reasonably unlikely that any random traffic will happen to match this particular pattern. It's possible there's some r

Re:

[nicolastheadept]

This is what happens at my College. After a certain time a box pops up saying it will shut down in five minutes unless you tell it no.

Re:Uh, Sleep Mode - shutdown?

[brain159]

The student who walked away from (or, fell asleep at) the PC without saving learns a valuable lesson.

Re:Uh, Sleep Mode - shutdown?

[operagost]

Or, you could just use the power management features present in every PC and OS since 1994 and have them go into standby or suspend.

Re:

[seaturnip]

Yes, a lesson about stupid IT departments and their apathy towards anybody getting work done.

Re:

[xelah]

You save documents less than every five minutes?? Wow! On a Windows PC?

I've long had a habit of pressing the save shortcut after every change I make. A sentence just doesn't feel 'finished' until I've pressed the save key...

Re:Uh, Sleep Mode - shutdown?

[tezbobobo]

At one school I administered (400+ PCs on the student network) we used a product called hadguard. It could remotely shut down computers (among many other AMAZING things) and group them how you wanted. You could shutdown, boot up, restart a parrticular classroom. This was only the icing of the cake. HDGuard is completely amazing and I recommend it to everyone. Besides purchasing it once I am in no way affiliated with the product.

Re:

[Anonymous Coward]

Quite a few of us like to work in the night (from 2 pm onwards)- mostly because our best working hours are during that time. Also it is a lot more helpful if you have a geographically distributed team.

Re:Aside from being green...

[Wordsmith]

I'm against anyone working after 2 p.m., myself.

Re:

[Nelson]

Does this seem like a really obvious extension to switches? Why not power down ports and then power them on via 802.1x requests. Kind of hits two birds with one stone.

viruses

[dsanfte]

The correct plural of virus is viruses. [wikipedia.org]

Re:

[Anonymous Coward]

Because "virii" is retarded from any angle. If you want to make up a "smart" sounding word use "viri" instead. It's still wrong, but it's less idiotically wrong than "virii". "Viri" at least has some logic to it, like "Vaxen". Imagine if some idiots had pluralized "Vax" as "Vaaxen". It's wrong and makes no sense.

Re:viruses

[Fred_A]

I think he should go the whole way and call them viriises. It's technology after all.

Re:Create job to force automatic reboot or shutdow

[dustmite]

Someone might be using it ... there are always a few late users. Trying to determine if a computer is in use in order to shut it down isn't always that simple.

I suggest a simpler, low-tech solution - just stick up visible signs in the labs, and on some of the major office floors, asking people to shut down the computers in the evenings ... it won't be a 100% solution, but most people would probably comply, so for the comparatively little effort put in I bet you can hardly get a better return.

Just the energy savings on that many computers would be not insignificant.

Re:Create job to force automatic reboot or shutdow

[CastrTroy]

Microsoft seems to have figured it out. They shut down my computer for me quite often to do updates.

Re:Create job to force automatic reboot or shutdow

[PFI_Optix]

At the school I work at, we have an automatic shutdown at 6 PM. It has a five minute timer and is preceeded by a text file in a DOS window reminding people that there is an "ABORT SHUTDOWN" option in their start menu if they are using the PC and the shutdown process begins.

Two simple batch files for XP, on in the All Users startup directory, one in the All Users\Information Services directory of the start menu.

Startup:

AT /DELETE ALL (or whatever the syntax is) - to prevent the AT table from getting crowded with dozens of the same command
AT 18:00 "shutdown -t 600"

Abort:

Shutdown -a

We reset the AT table every day just in case some know-it-all high school student finds out such a thing exists and starts screwing with it. For the most part, though, not even the techs knew such a thing existed until I proposed using it.

We tried a lot of other ideas, but this is the simplest and most user-friendly. Big signs don't work, teachers and lab aids are no better than the students about following directions. Since implementing it 18 months ago, we've gone from having roughly 900 PCs online at night to about 100...including servers, timeclock systems running thinstation terminal sessions, and technology and admin workstations that are excepted from the shutdown policy.

Re:

[PFI_Optix]

Too bad for them. People leave themselves logged in all the time at the end of the day. Too many people are sloppy for us to make allowances for things like that.

It should be noted that this is a public K-12 school, not a university. There aren't many people around at 6:00 outside of administration personnel. Those who are have been warned what will happen; if they get up at 5:58 to do whatever and come back seven minutes later to their PC shut down, they'd better have saved their work.

Re:

[SkyDude]

It should be noted that this is a public K-12 school

If I were a taxpayer in your district, I would appreciate the savings on the school's electric bill.

Re:Create job to force automatic reboot or shutdow

[camperdave]

In a corporate environment, that sort of decision is made at a corporate policy level, and not at an IT personnel level. Nobody would get fired.

Re:

[PFI_Optix]

They really should get their own computer instead of sneaking into a public school. Do you have any idea what kind of trouble that can get you into these days?

Re:

[gbjbaanb]

because we're the ones who frequently have to drop what we're doing quickly and might forget to save an open document. We're also the ones most likely to work odd hours.

yeah, just like the councillors who are saying we need to get rid of parking spaces at workplaces, but not at the council offices as he needs to get to his constituency at a moment's notice.

You're focussed on yourself, I bet if you went out and asked the users you'd get a lot of replies around "yeah, the stupid thing tries to shut itself off just when you don't want it to".

Live by your own rules.

Re:Create job to force automatic reboot or shutdow

[ATMD]

Why not just have the machine shut down automatically on logout if it's between two specified times (eg. 9pm and 8am)?

Re:

[Tanktalus]

Seriously. You need to take a couple of simple electrical courses. Specifically, volts-for-dolts a physics course on magnetism, and a practical, hands-on course working with electric motors. All handily part of an electrical engineering degree. When there is load on an electric generator, it takes more energy to keep it moving. Of course, we need to keep it moving at the same speed (60Hz or 50Hz, depending on which part of the world you're in) at all times, so that means using more coal, gas, whatever,

Re:

[Single GNU Theory]

The net result of this: there is a lot of electricity available at night. Large businesses pay way more for electricity at peak. Off-peak loads aren't even measured some times

You may find this interesting. I used to work at a nuclear power plant. The cooling reservoir was connected via a spillway to a nearby river. To cover peak loads, there was a hydroelectric plant on the spillway. At night, the power company I worked for bought tons of cheap electricity from another utility to run the hydro plant b

Re:Aside from being green... Just let them stay on

[hjf]

so, driving uphill uses the same amount of gasoline that driving downhill?

the power station does throttle down at night. they keep the generator at the same speed (3600RPM I guess, to give you 60Hz). but they don't need the same amount of fuel to keep it going. the usage on the grid acts like a brake on the generator, in the same way that the road conditions affect your bicycle.

if it's steam-based (gas, coal, nuclear), you need more steam to keep a higher pressure, to keep the generator rotating at the same speed, and that means heating more water, and more water needs more energy, and more energy needs more fuel. hydroelectric plants shut down unused turbines.

Re:

[NeverVotedBush]

It's amazing how people can have such strong feelings about an issue and have them be based on such utter nonsense.

Have you ever jump-started a car? Did you not notice how when you connect jumper cables to the vehicle with the dead battery, the running car has its engine get slogged down by the extra load? And does it not stand to reason that if you want to hold the engine at the same RPM (as an electric utility has to do to hold line frequency) you have to feed the engine more gas to do so?

If you dra

Re:

[dknj]

You run these simulations on Becky's computer over in the Human Resources department too? That's efficient computing. But I can bet you $100 that Becky's computer is sitting at a windows logon screen at 2am with 0% cpu activity. Why not power down the machine automagically at night and boot it back up with Wake On Lan in the morning?

Imagine if even 5% of computers in the US did this, we'd drop our carbon footprint drastically.

We power down at weekends

[Realistic_Dragon]

During the week machines are left up to push automatic updates (5 minutes of downtime, times 10k employees, is about $80,000 of billable time). At weekends they get shut off either manually or under remote control.

Re:We power down at weekends

[Albanach]

What's the cost of 10k * 261 days * 12 hours of power?

Surely you could use wake on lan to wake the machines then do your rollout 10 minutes later? Or do a patch install when the machine is turned on and connects to the domain controller?

In windows I'm sure you can set the time between warning appearing and shutdown ocuring. Give 600 seconds warning and you could probably shutdown 90% of the machines overnight.

Re:We power down at weekends

[Ironsides]

What's the cost of 10k * 261 days * 12 hours of power?

Well over half a million dollars if I did the math right.

Surely you could use wake on lan to wake the machines then do your rollout 10 minutes later? Or do a patch install when the machine is turned on and connects to the domain controller?

Unfortunately, this doesn't always work well. On some networks, the machines will auto-start up the moment they receive a packet, even if it isn't intended for them.

Re:

[ACMENEWSLLC]

We ask our users to reboot their PC & not to shut them down for the patch reason. Unlike most, we keep up to date with patches. We watch Secunia and automatically roll out kill bits for insecure active x controls, automatically patch Windows via WSUS and all the other pieces of software such as Adobe's, Real, Apple, et al. That is a lot of software.

We do this automatically eg automated.. If we do it while the non-admin user is signed on, many of these packages fail to install. Flash and others

Re:We power down at weekends

[jimicus]

In windows I'm sure you can set the time between warning appearing and shutdown ocuring. Give 600 seconds warning and you could probably shutdown 90% of the machines overnight.

You're assuming that 100% of machines in use are doing something interactive (and therefore have someone sat at them). This is frequently not the case.

Probably not that much

[Sycraft-fu]

Computers don't use a whole lot when they idle. Unless you are loading them up with lots of drives and a big GPU, you'll probably find they draw in the realm of 50 watts just idling. Ok so looking at my power bill I get about $0.06 per kilowatt hour of energy. At 50 watts of draw it takes 20 hours for a computer to use a kWh. Running the numbers I come up with about $100,000. So assuming the costs quoted by the grand parent are correct, one patch time would just about pay the power bill the whole year and t

There's also three other reasons to do it

[Sycraft-fu]

One is efficiency. Powersupplies, good ones at least, are quite reliably most efficient at half their rated power. So get a 600 watt power supply, it's efficiency peak will be at 300 watts. Hence it behoves you to buy a powersupply that's double what you need. While the difference generally isn't major (maybe 2-4% between half and full load) if you are spending the money on a quality, efficient supply anyhow, why not get one that runs at its most efficient point?

Another along those lines is if you want a go

Re:We power down at weekends

[Eivind]

14 hours a day times 52 times 5 times average power-consumption for an idling desktop times 10K employees works out to:

14*52*5*0.15*10000 = 5.460.000 Kwh of electric power, WORSE in the time when you use AC, because you'll need additional AC to get rid of that extra heat.

With average power-prices of 10 cent, you'll spend more than half a million dollar just paying for the electricity, in practice with AC and all you'll probably pay a million.

So, you saved $80K and wasted a million. Way to go !

Common wisdom

[Improv]

It's probably smarter to track IP addresses unless you control all the switches :)

Common wisdom (which may or may not be actual wisdom) suggests that powering up/down of computer power supplies is one of the largest sources of "wear" on computers nowadays, and so it's best to avoid that (replacing system components and increased costs in the industries to make this possible should be factored into eco-costs as well). Having systems go to sleep to various degrees presumably gets one much of the way towards b

Re:Common wisdom

[Planesdragon]

Common wisdom (which may or may not be actual wisdom) suggests that powering up/down of computer power supplies is one of the largest sources of "wear" on computers nowadays, and so it's best to avoid that

Nope. That would only be true if all of the three are true:

1: Power-cycling actually reduces the MTBF opposed to just leaving it on.

2: The reduced MTBF is lower than your company intends to keep the asset.

3: Cost-savings from the "increased" MTBF by leaving it on is greater than the electricity (+ increased A/C cost) cost to run those 300W power supplies all the time.

Of the ~6 computers I've had to failure, they all lasted far longer than even a five-year technology plan, AND did not fail due to simple wear and tear on the circuits. My anecdote isn't data, but it does make me question your conventional wisdom. (Especially since those PCs I know that are left on all the time don't have a significantly increased lifespan.)

Re:Common wisdom

[ScrewMaster]

Most failures of any electrical or electronic system occur during startup. That's when subsystems haven't fully stabilized and experience high inrush currents, with concomitant spike heating and other stresses. It's the same reason incandescent lights usually pop when switched on, but rarely fail when left lit.

I never switch any of my systems off, and failures are extremely rare. I have all monitors and flat panels automatically power down, but I leave hard drives running continuously. About the only time I have to replace something is when I upgrade every few years. Yes, it adds a few dollars to my electric bill, but I save in other areas there, and it is worth the peace of mind.

Even fans (which are the weak link in most PCs) can run for ages if you spend the money to buy quality parts. It helps to have a good HEPA filter in your computer room, and keep the machines off the floor. Fans last a long time without dust in the bearings, and a dust-free computer runs cooler as well.

Re:Common wisdom

[toddestan]

I find that powering on a system is an excellent time to catch things that are starting to fail. I'd much rather catch the harddisk when it's starting to have issues spinning up and replace it then, as opposed to finding out the harddrive is a brick when I try to start up a computer that was just shut down after months of being on. It's been a long time since I've lost any data on a harddisk that was regularly powered down. Furthermore, I've also found that drives that are only run a few hours a day outlast drives that are run 24/7 anyway.

Re:

[oldhack]

Common wisdom (which may or may not be actual wisdom) suggests that powering up/down of computer power supplies is one of the largest sources of "wear" on computers nowadays, and so it's best to avoid that (replacing system components and increased costs in the industries to make this possible should be factored into eco-costs as well). Having systems go to sleep to various degrees presumably gets one much of the way towards being more eco-friendly without so much of this wear. That said, presumably a rigou

Re:Common wisdom

[Improv]

False certainty characterises a bit too much of dialogue on the internet, I think, and makes it hard to be careful and humble in discussion. If I felt more sure of what I were saying, I'd have phrased it differently :) Presumably :P

Preventing Infection?

[TheRaven64]

A lot of places require machines to be on overnight because that is when automated update, monitoring and scanning tasks can run without impacting users. Of course, the machine could be configured to automatically shut down when this is finished. Actually shutting down is typically highly inconvenient since the machine loses state due to 30 years of bad OS design when this happens but a suspend-to-disk mode is a viable alternative.

Re:

[saarbruck]

on a related note, my current employer requires so much intrusive anti-virus, anti-malware, update & corporate IT policy scanning software that for the first 1-2 hours after boot up, the PC is next to useless (which has always seemed weird to me: if the PC was clean when it was turned off, why does it need a full scan immediately after booting?) Anyway, I honestly do feel bad for all the baby salmon I'm killing by leaving my rig on overnight. Somehow I don't see my boss buying the excuse that I'm not

Re:

[InvalidError]

In hybrid (Windows-*nix) environments, logging off, suspend or hibernate means terminating all networked application (X-Windows and Xterm) sessions and having to waste 20 minutes each morning to re-login on the hosts and re-open everything. With even junior engineer wages being $20+/hour, those minutes can easily cost over $5 each day and make spending $1/day on the power bill save $4/day on the bottom line.

Every power or network outage and mandatory-reboot Windows update, even overnight, usually causes maj

Good idea!

[cromar]

I would go with a reward plan. You could do something like give the top three most energy efficient people a gift certificate to the campus eatery (or whatever really). Calculate how much money is saved (out of everyone participating) and use part of that money to create a pool for the prizes. (It seems like for a large enough group of people, the energy and maintenance costs would reduce considerably, but I wouldn't really know ;) I know I would definitely turn off my work PC every night if I got a free lunch!

Re:Good idea!

[fotoflo]

The top three "energy efficiency" people are the people taht never show up for work and leave their computers off all the time.

Re:Good idea!

[Angostura]

Hmmm, lets see if a small amendment to the metric could circumvent that problem.... no, you're right it's an impossibly difficult flaw to fix.

Hibernate

[TheMeuge]

It's time that all large campuses configured their systems hibernate automatically, if left unused for 30 minutes.

Really, there is no reason NOT to use the power management settings built into the OS.

Re:Hibernate

[superflit]

I did a project in My Campus with more than 8000 desktop computers.
It saved something like 33% power consumption (measured, before and after).
after midnight all desktops that are not in the excluded list hibernate automatically.
I used python + MFC . Was very easy and simple.
It is time for the Sysadmins start to program and make better use from the technology (not just, next-next-finish)..

And I didn't receive any raise besides saving a lot of money to University.
Shameless promotion: Looking for a new job in developed country.

Re:

[ScrewMaster]

I'm sure he'd be happy to after you give him a job in a developed country.

This [mvps.org] was the top result in Google when I searched for "windows remote power down API".

Re:

[superflit]

Paid 50% in taxes, escaped from robbers.
Twice threaten by fire guns.
Being honest when everyone is "taking a good time"..
Voted, volunteer work..Teaching..

We can change places..

I dream when I may take a dinner and walk some quarters by night without fear..

Re:Hibernate

[The Second Horseman]

On a Windows XP system, you also want to set the CPU performance in the default power profile to "ADAPTIVE". I'd actually think you'd do well to set the hard drives to spin down and the monitor to turn off after 15 or 20 minutes, set the system to suspend after 30 or 45 minutes, and hibernate after an hour and a half to two hours. You might have to exempt some systems from hibernating - some software and drivers don't always react well to hibernate, and it would be a pain in the (*#)(@ to have to restart after lunch or every meeting. Suspend is a good middle ground. With something more disruptive, a company could well look at that and say "it's not worth the few minutes per day of productivity loss, when factored against the employee's salary + benefits cost." Especially if it leads to calls to your internal helpdesk to try to recover documents in progress or some other work. By the way, productivity vs. conservation is one of the reasons organizations need to be given incentives to conserve power if we want them to do it before energy prices actually exceed cost per hour of labor.

Re:

[AmiMoJo]

"Stand by" in Windows is actually quite a good option. Most PCs take 5-10 seconds max to come out of it, yet only use a few watts while sleeping. The computer can wake itself up during the night for updates etc.

Just remember to stagger start up on the machines, or you might trip a breaker.

Spinning down HDDs is not a good idea, at least on Windows. It tends to spin them back up after 1 minute anyway, causing constant stop/start cycling.

Re:Hibernate

[Atlantis-Rising]

Obviously, he works for the NSA.

Do you support wake-on-lan?

[rmcd]

I'm at a university and many of my colleagues leave their machines on overnight because they sometimes need access to their machine, either to retrieve a file or to run a program. If the IT folks provided everyone with a wake-on-lan script then everyone could turn off their machine. For years this has seemed to me like a no-brainer.

Is there some security or other downside I'm not aware of? Is WOL not reliable?

Re:Do you support wake-on-lan?

[arth1]

WOL takes longer than the timeout period for many services (including, but not limited to WINS, and some windows networking components will take up to three quarters of an hour to recover if a service is initially reported as unavailable while the machine appears to be up).

Another concern is whether your servers are up to handling all the PCs coming on at once in the morning. People leaving Outlook running at all times is actually a Good Thing for IT, cause the alternative of thousands of people hitting the Exchange servers at the same minute would kneel even the biggest distributed servers. Then there's similar concerns for the domain controllers, DHCP servers, proxy servers, or you have it. Leaving a substantial part of the machine park already logged in can save a lot of hardware and configuring.
If shutting everything down, at least a staggered start-up could be prudent.

Re:Do you support wake-on-lan?

[QuantumRiff]

Why on earth would you need to access your machine to get your data? Are you actually storing important data on the desktop? You really, really need to look into File Redirection in GPO's. We move desktops, Application data, and MY documents to a network drive, that is actually backed up every night. Users don't have to worry about losing data because their drive dies, or whatever.. They can also move to any other computer, and have almost all their apps running on it. (there are a few exceptions for specialized software) On our student network, we setup every desktop to power down at midnight. All run virus scan's updates, etc, between 10pm and midnight. (labs close at 10pm.) The servers stay on, so files can be reached remotely. In the morning, only a few machines will automatically turn on, most wait for someone to push the button. The power saving for us were significant enough to not worry about a student having to wait 30 seconds for a machine to boot.

I'm going to roll this out to our admin network computers as well. We are really saving noticable amounts of money, because not only are the machines not powered, but the AC doesn't have to run to keep the rooms cooled. THe only glitch I have ran into is when I need to push out updates to all computers, and some were not turned on that day. In the late afternoon, I use WOL to wake up all computers on campus.

Why power down?

[lukas84]

We don't completety power down any of our desktop machines. Users log off in the evening, and machines go to standby/hibernate after enough time has elapsed. Thus, users do not have to wait in the morning till the machine boots.

Machines are woken from sleep to deploy updates, etc. Many of our desktops are able to accumulate 30 days of uptime before the next patchday.

Energy consumption is a non-issue. We don't pay much for electricity.

The rest of the infrastructure - printers, faxes, access points, etc. runs

Re:Why power down?

[mollymoo]

Energy consumption is a non-issue. We don't pay much for electricity.

Reducing energy consumption isn't just about saving money, it's about not fucking up the planet too.

Re:Why power down?

[Creepy Crawler]

Unfortunately, "fucking up the planet" as you so eloquently put it isn't in the books as a cost.

Re:Why power down?

[NeverVotedBush]

I'm sorry you're not too.

News for ya...

[Otter]

My Health Sciences Campus has about 8,000 desktop computers, and on any given night about half of them are left on.

"Health Sciences Campus" sounds like at least a few hundred of those are grad students and postdocs chained to their desks by their PIs...

I'm not sure whether your definition of "powering-down" includes sleep; it seems like reasonable default (or unchangeable) power settings should be adequate to address your concerns. Admittedly, that's easier done in a company than in the free-for-all of aca

scripted per department

[justanotherlinuxguy]

I did this a while back, the trick is some machines are 24/7 others are 9-5 ers. I coordinated with dept heads to identify what entire departments could be shut down then scripted a prompt to fire at 7:00pm to look for any user feedback, working late crowd, then 30 minutes later do a shutdown if no response was received. This took care of most machines. I never got to the mixed departments, greener pastures called.

Ping?

[pete-classic]

Can you ping those machines? They may be sleeping but powering the NIC for WoL. That leaves them drawing very little power and immune to any IP-based attacks.

-Peter

Increased probability of HDD failure

[methamorph]

Shutting down at the end of the day and powering up the next morning increases the probability of HDD failure. It's better for the HDD to run all the time than to cold boot every morning.

Re:

[ditoa]

We have had the exact opposite of this after implementing a shut down at night policy.

Re:Increased probability of HDD failure

[toddestan]

Harddrives are mechanical devices, and are wearing out anytime they are powered up and running. While I'm sure that a drive does get stressed a bit more when it is turned on, I can guarantee you that a drive that runs for 40-50 hours a week is going to last longer than a drive that runs 168 hours a week.

There are also other benefits. A harddrive that has motors or bearings that are starting to fail can be caught when they have trouble spinning up and be replaced before they totally fail, preventing data loss. Furthermore, if a head crash occurs when no one is around (during the night or the weekend), having the the heads banging and grinding against the platters for hours or days is really going to hamper any recovery efforts.

We have power down at night policy

[ditoa]

I work for a large blue chip company and we have a strict policy of powering down at night (including monitor). We regularly audit the records to ensure the machine is powered down and users who are not are requested to always remember. A few users take a few reminders in order to do so and I have heard every excuse under for why they left it on and while some are valid the majority (95%) are not. Our reasons for pushing this policy is purely to save money and reduce unnecessary running time of the equipment. However we are in a position where only laptops users have VPN access so if they need to login to the network from home they already have their laptop with them. If we had open VPN access to desktop users I am sure we would see a lot of users leaving their computer on so that they can RDP into it over VPN.

It took about 6 months before we were at a realistic level. We have 633 desktops on our site so there is normally always a valid reason for one or two to be left on (valid reasons being batch copy, verify or processing of files). For those interested we have had a reduction in the amount of equipment failure (HDD mainly) as well as pretty good cost savings for power. Not to mention running greener (which regardless of if you believe in global warming or not is good).

Re:

[cnettel]

Just a thought: the (supposed) increased failure rate for HDDs wouldn't come within 6 months. During the first period, it's instead perfectly reasonable that the reduced number of power-on hours decreases the calendar-based failure rate. The interesting issue is whether your HDD failure rates increases significantly within a 1-3 year timeframe.

Re:We have power down at night policy

[ditoa]

Sorry I should have clarified. It took 6 months to implement but we have had the policy for a little over 4 years now.

Power vs. operational

[Todd Knarr]

I tend to leave the computer on overnight, but with things like monitor power-down and CPU idling enabled. When it's not doing anything it drops about 90% of it's power consumption after 15 minutes, and even when working with the monitor off (eg. running the nightly backup) it's still running at less than 50% of full power. If I power it off, by comparison, it can't run it's virus scan, backup, update check and the like overnight and has to do those things while I'm trying to use it during the day. Plus there's wear and tear to consider, I've noticed that the office computers that get turned off and on every day tend to fail and need replacing several times before mine (that stays on all the time) has a failure.

So my preference is to leave computers running but with power-saving features set to minimize power without shutting things down. This means hard drives continue to spin but the CPU goes into low-power idle mode. The monitor goes to suspend mode (beam and deflection power is off but the circuits and coils are kept warm), not powered-down completely. That seems to be the best balance between reducing power consumption, allowing it to run maintenance operations overnight and minimizing wear and tear and thermal stress on the components. If management absolutely insists on ignoring those last two in favor of the first, wake-on-LAN is essential to allow nightly maintenance to happen.

Re:

[fermion]

I suspect that almost no machine is used 100% during the work day. I would suspect that most machines would be better off set to automatically sleep after a period of inactivity rather than leaving it on all day, turning it off at night, and then turning it back on the next day.

The idea of turning off a machine is an old and out of date idea. Power management build into machines is now quite good. Another consideration is that commercial machines, at least, hit a central server on startup, and if every

Re:Power vs. operational

[edunbar93]

Actually, leaving your hard drives on is the biggest source of idle-computer power drain. You can also set Windows to turn them off after a set period of time, say an hour of inactivity. It doesn't take long to spin them up anyway. Or you can set it to go into sleep mode, which does the same thing and more.

Also, for the love of god, get an LCD. Modern LCDs are leaps and bounds better than CRTs in every way, especially power consumption. And they're dirt cheap too.

Re:Power vs. operational

[toddestan]

Also, for the love of god, get an LCD. Modern LCDs are leaps and bounds better than CRTs in every way, especially power consumption. And they're dirt cheap too.

So long as that CRT still works, it's better for the environment just to keep using it. The added electricity usage is far less than the energy and environmental costs of properly disposing of that CRT monitor, not to mention the environmental and energy costs of producing the replacement LCD.

Re:Power vs. operational

[Todd Knarr]

Your point #1 is false, I'm afraid. Yes, the components were designed to work over a wide temperature range. The problem is the change. Gradual or not, components made of different materials expand and contract at different rates as temperature changes. As long as the temperature stays steady, regardless of what that steady temperature is, there's no problem. But if it's changing, even slowly, then the different materials want to change size relative to each other. That produces stress. You can see this in the thermostat that controls your home's heating and cooling. It's made of strips of 2 metals with drastically different thermal expansion ratios bonded together. As the temperature changes, the bimetallic strip physically warps. The same thing happens inside every chip and solder joint in your computer every time it warms up or cools down. And eventually those joins start to crack apart. That does two things. First, it increases the electrical resistance of the joint. Higher resistance = more heat. Second, it disrupts the thermal transfer from the chips themselves into their carriers by disrupting the tight physical contact between them. So now the chips aren't being cooled quite as well, and at the same time they're running hotter internally. All that turns into a shorter lifespan.

As for #2, not anymore. That surge effect's confined to the power supply itself. It still affects the PSU, but no worse than normal power-line noise will. The big effect is actually on the drive motors and bearings. Stopped, the bearings settle a bit. When the platters are spun up, it takes a moment for the bearings to lift up on their film of lubricant again. One cycle like that puts as much wear on the bearings as many hours of steady spinning. And the power cost of keeping a drive spinning is minimal. Think about spinning a wheel, which is all the drive's platters are. It takes a fair amount of effort to start a heavy wheel spinning and get it up to speed, but once it's going it takes very little effort to keep it spinning at that speed. It's enough on my large drives that I can actually see the spike in power draw at the wall socket when a drive is spun up from a dead stop.

When people comment about how their machines aren't having problems, I'm often a bit skeptical. I've had quite a few people ask me for help with a computer that "just started having problems a few days ago". When I throw my hardware diagnostic program at it, I get a veritable flood of red malfunction and error indications. On the last one, "Every now and then it won't boot, and occasionally a program dies for no reason." turned into having to replace the motherboard and all the RAM (every stick of memory had at least 2 bad columns, and the EIDE controller was fried and corrupting data during large DMA transfers) and restore the entire system from original media and backups (losing about 10% of the data in the process, the backups didn't go back far enough to contain uncorrupted copies of the files).

I used to turn my machine off at night ...

[LaughingCoder]

... but now IT has loaded so much crap on it ("desktop agents" [ie apps that spy on me], antivirus, patches, etc) that it is fully 15-20 minutes after turning it on before it is usable. So now I never turn it off. I did the hibernation thing for awhile, but then it stopped working for some reason and I haven't been able to fix it. And if I ask IT to fix it, their solution is always the same for every problem - wipe the machine - a tad inconvenient for me, but pretty efficient for them I suppose. Sigh.

Re:I used to turn my machine off at night ...

[_xeno_]

Me too!

In any cases, I always leave my computer (a laptop) on during the week. I shut it off on weekends, but due to the software inventory tracker and the required anti-virus scans, I always leave the machine on during the week so that I can actually use it during the day.

The real problem is that the anti-virus scan is so slow that it takes a good three hours. The inventory scan is somewhat better, and only takes about an hour. In both cases, the machine drags to near unusable levels while the scan is running. Given that it's a dual-core machine, this is really a testament to just how screwed up Window's I/O scheduling is - both involve lots of file reads, which apparently causes Windows to drag to a crawl.

Not to mention that hibernate and to a lesser degree suspend appear to not work well with certain drivers on my system. Using hibernate kills the wireless drivers, which isn't a horribly big deal when I can physically plug the system in but it does mean that I just shut the thing off when roaming about, since I'll have to reboot anyway.

But it's that three-hour IT required virus scan that keeps me leaving the machine running nights. That's a real productivity killer during the day. Fortunately it's only scheduled to run once a week.

The inventory app, on the other hand, runs daily for some reason.

Why not use standby?

[schnikies79]

S3 standby can be woke remotely if need be, and on new computers it only takes a few watts.

Wake on LAN

[necro81]

It shouldn't be necessary to leave machines up and running all the time just to be able to push updates or run virus scans (etc., etc.). That's what Wake-On-LAN [wikipedia.org] is for. It allows you to put a computer to sleep, then wake it remotely when it is specifically addressed to (say, by a remote administrator). Sure, it needs to be hardware supported on the motherboard, and the BIOS needs to be set up just so, but it isn't that hard to implement on a new machine that you need to configure for the company anyway.

I have been told they turn off overnight in Japan

[hax4bux]

Among other projects, I worked on the power supply controls for the Cray Super Dragon. No, you probably never heard of it, but it became the Sun ES-10K.

This box had variable voltage power supples which required me to adjust them from cold start. I had to calibrate A/D, take samples, tweak, etc all through JTAG and cumulatively it was quite slow. Like over an hour.

My manager was not impressed, I shrugged and said "who turns these off?" - and the marketing droid/product manager said "they do in Japan". Fine. The hardware people were nice enough to give me multiple JTAG lines and power up time shrank to acceptable limits.

I have never been certain if this was a "Spinal Tap" riff or it was really true.

Studies?

[stormguard2099]

Does anyone know of any studies about HDD and other hardware failure rates when powering down as opposed to sleep/hibernation/etc? It seems like everyone has their own antecdotal evidence but I haven't seen anyone show any kind of proof beyond that about what is actually best for the hardware.

Waste not, want not..

[Ancil]

I worked for a very large (top-3) pharmaceutical for years. They always asked employees to shut off their computers at night when they went home.

Then one day, they sent out a campus-wide email telling people to leave their computers on all night and over the weekend. They used the CPU cycles to run high-performance scientific computing jobs, saving the cost of buying a supercomputer.

Of course, not every company has a need for spare CPU cycles. This place did a lot of protein-shape searches etc..

workstations by day, cluster by night

[nategoose]

The CS department at the college I went to used to turn off all the PCs at night but now has them set up to start doing scientific calculations during the times when the labs are closed. They use power during this time, but it's not wasted.

Re:workstations by day, cluster by night

[anno1602]

Well, in a university, "scientific calculations" could mean actually running simulations/models/whatever, that is, things for which you would normally use some super computer. So it is not necessarily a question of "getting away with this", it might make perfect operational sense and boost the uni's available computational power and thus productivity. Depending on what industry you work in, you could do a similar thing in a company, for example a nightly distributed build-and-extensive-regression-test in a software shop, and save on purchasing and operating dedicated computers. There are lots of quite productive uses for distributed computing. You just need to think about what your company might do with it. Of course, if there is no productive thing the computers could be doing, they are better turned off.

Yes, but not strictly enforced

[pla]

Do security concerns and power consumption issues matter enough to do this?

Yes and no.

When I first got comfortable in my current job, I made a big push toward "greening" our IT resources. As one obvious (erroneously, as I'll explain in a sec) step in this, I convinced most of my users to shut down at night. If we need to push out updates, WOL works just fine for turning machines on a couple hours before the start of the day, and it doesn't impact anyone during working hours.

Then I learned how electric billing actually works for commercial users - Put simply, your company doesn't care if machines stay on all night, because they pay based on their peak load, which will always occur during normal business hours. I had applied ideas that make perfect sense at home, to an environment where they don't apply.

Now, that doesn't mean we should just leave machines on 24/7 - Using electricity has an an environmental aspect in addition to the monetary cost. But if it inconveniences users by more than a few seconds every day, any conservation efforts will actually cost the company money in the long run.


So, I still encourage my users to shut down, and 95% comply. But if they consider it too much of a hassle, I can't financially justify forcing them to spend the first minute of the work day waiting for their machine to boot (not that anyone really works for the first five to ten minutes of the day, between coffee, hitting the bathroom, and just getting the obligatory morning socializing out of the way).

As for the security aspect of this, the servers must run 24/7, and any attacker would target them rather than some random user's desktop. I don't worry about an attacker using a compromised desktop as an intermediate step to the servers, because the desktops have no more privileges on them than anything else inside the firewall (and even then, not much more than a totally untrusted source, except for nonconfidential shared resources that we could restore in a matter of minutes if necessary).

Overnight tasks

[kylegordon]

I can't help but laugh at those that quote reasons such as 'automatic updates' and 'antivirus scans' as legitimate reasons for leaving a computer on overnight.

With many enterprise management tools, such as Zenworks, it's quite simple to schedule a wake-on-lan task to wake computers up at say, 6am, to perform their daily tasks. It can even be configured to push out an automatic reimage of the machine. Once the updates and scans are done by 7am, people are just beginning to come into the office, yet you've still had a whole 10 hours of downtime. Incidentally, I've not seen a single computer in the past 4 years that doesn't support WoL on the mainboard NIC. Big bucks enterprise manglement apps aren't even required. A simple cron job, and some wakelan/ether-wake/wakeonlan/Net::Wake magic will do it for free. Just gather a list of Mac addresses with ettercap or your friendly ARP table or asset management app/spreadsheet.

May will say that the bandwidth requirements of updates squeezed into the 6am to 7am slot will degrade systems, but that's where a background process such as BITS [wikipedia.org] should be used (as demonstrated by Eve Online, Zenworks, Microsoft and Google). The virus updates are a minor bandwidth requirement if you have suitable leaf services, and the actual scan is only locally intensive.

Being a public sector organisation, we're working towards a greener profile (due to govt policies), and all the tools are there and working. It just needs some effort on the part of the administrators.

Tool to do this The Right Way from the EPA

[hklingon]

For some odd reason windows stores the Power Management stuff in the registry in a Binary (!) obscure (!!) machine/driver-specific acpi (!!!) way. This means doing stuff with it via group policy is tricky at best. Fortunately, the EPA has a really great solution we've been using for years and is absolutely fantastic.

Unfortunately the EPA's EZ GPO page seems to have gone poof or something recently, but you can get it here. [terranovum.com]

Basically, you push a (simple) msi to the machines (I do this a lot of the time via psexec [microsoft.com] (props to Mark Russinovich) but there are other methods. Once you have that running on the machine you can configure how you want your machines to behave/re power management:

• Monitor Sleep time when logged in
• ...when not logged in
• Hybernate or Suspend to ram options
• Allow logged-in users to override (e.g. laptops/presentation mode)
• Non Intrusive setup/no options

We also have a script that runs at midnight a few days of the month that does the magic packet thing as has been mentioned so WSUS and/or SMS (or SC:CM) can do their thing and automatic updates run as normal. In a few "why does my machine have to boot up every day this sucks" user groups we have a scheduled job to send magic packets about 15 minutes before they arrive to wake up their machines. With hybernate they hardly know anything happened.

Some scientists need their computers on

[Werthless5]

I know in particle physics we need to leave our computers on overnight quite regularly. We share computing resources and often run simulations for several days (or longer). Shutting down the routers and switches connecting one computer to the rest of the particle computers in the building effectively cancels the simulation since huge datasets might be spread across 7 or 8 computers. At CERN, when the LHC turns on there will be thousands of computers running 24 hours a day for many years. At a university, obtaining your sample set of data may require at least a day (you're expected to pull the data and then work with it rather than using CERN computing resources, although the specifics haven't been worked out yet). Some projects just require that much time and energy. Most days you should be able to shut off large portions of the network, though.

I'm certain there are other sciences that have similar concerns. I think the best way is to send out a friendly e-mail reminding people to turn off their computers when they leave. That should get at least a handful of computers off for the night. Depending on how successful or unsuccessful that strategy is, shutting off computers that are definitely unnecessary (public access terminals for example) would be a fine idea.

Personal experience with 3300 PCs

[thalassinos]

I work for a medium size European bank. Total workstations aprx. 22000 in 13 different countries.

We used to leave all our PCs on all the time in order to run updates, patches etc.

In my area of operations there are only about 3300 PCs. Nine months ago we implemented a policy where all users were required to turn off their PCs (not servers) at the end of day. Wake-on-LAN was used to turn the PCs on during the night for updates and 15 minutes before the start of the workday.

Very conservatively, we estimate that we will save about EUR153000 (USD225000) every year (I live in a country with very high electricity rates).

So, it is definitely worth it financially, our users were not adversely affected at all and it helped morale by making the workplace a greener place.

Wow!

[soccerisgod]

In the company I currently work for, every PC is powered down at night, with only a few exceptions. There are obviously servers still running, but they are actually doing something (backups for instance), and there's sometimes a few machines involved in over-night test procedures. Frankly there is no valid reason for keeping a PC running if it isn't being used. Same goes for the home PC, btw. Switch it off at night.

Answer: force them to power down

[buss_error]

Our AD policy (since most computers are Windows) forces a powerdown at 7pm (our offices all close at 4:45pm, except for a few.) The user can abort the shutdown by clicking on a button, or can simply reboot. AD policy also exempts the systems we know shouldn't be shutdown (24 hour serivce points.) At this point, we estimate we save about $30,000USD per year in power costs for the 1/3 that have this impliments. (It's a big network.)

Interestingly, our network guys are having trouble routing wake on lan packets accross subnets, so we are looking at a T104 form factor linux appliences with multiple nics to send out WOL commands. Not sure this isn't a brain fart on the part of the network guys, or simply a limitation on how WOL works. Since we have other reasons for wanting a boxen on each network, it's a good excuse.

Why bother "encouraging" them? Force them!

[brunes69]

You are the sys admin are you not? Maintaining the operation of the network is YOUR responsibility. That includes maintaining it's carbon footprint.

Push out an update to disable screen savers, turn monitors off after 15 minutes of inactivity, and hibernate after 1.5 hours of inactivity (this saves them from having to boot up after lunch), and set Windows to require administrative privileges to change power management settings.

Piece of cake. If they run Linux s/Windows/Linux above.

Re:Easy fix

[UncleTogie]

You have to weigh that against lost productivity from employees needing to remote into their workstation at night and print that document to submit for tomorrow's 8:00am deadline. They'll now be able to blame IT, when the boneheaded logon-hours policy makes them fail.

You might be able to set up an "exception" ticket with the IT department, or set up a Magic Packet [wikipedia.org] arrangement tied to their machine.

Re:Easy fix

[smoker2]

You could also use that magic packet, to signal the computer to wake when the user swipes their access card first thing in the morning. By the time they reach their desk, their pc would be up and running. Ok, you have to link the access systems to a control server, but it wouldn't take too much hacking. They probably log accesses on a server anyway, so use that one.

Re:Easy fix

[hedwards]

One which is properly managed. It is a very bad thing when working nights and weekends is the thing to do. Except in a global situation where 2 in the morning at the work site might be noon where the worker is, or it the workers do most or all of their work at night. If it's just a matter of accessing some work files, that's what the fileserver is for, it makes far more sense to figure out how to set that up securely, than to hope that every one of a potentially huge number of computers is secure enough to access from offsite.

Encouraging people to pull all nighters or work on weekends smacks of an inefficient workforce or ineffective management. The only times people should be working those hours are because a job sprang up last minute due to an act of god or because the work requires that it be done when people aren't around. And in those cases if you've got on sight IT, it shouldn't be that difficult to set up an arrangement to cope with that. If you're going to have work done nights and weekends anyway, you may as well just outsource things to another timezone, and that's frequently a cost saving thing anyways.

Re:

[dal20402]

No, a properly managed company would allow users to work when they want to the greatest extent possible. Don't assume that everyone prefers the same hours you do.

And for most positions, there's no need to make each physical box remotely accessible to allow people to work whenever and wherever they like, just a remote home directory for each user.

Back on topic, I think users can be given some input; the key is to make them think about the issue. Run a script that asks each user when his or her PC should

Re:The Green fad, is just that... marketing nonsen

[KillerBob]

Obligatory disclaimer: This is my opinion, and may not reflect that of my employers. If you have a problem with it, take it up with me, not them.

I work for Dell. I can tell you for a fact that we take the environment seriously. The building I work in houses a 24/7 call center, but certain areas of the building are not 24/7. Corporate sales for the country are here, and take up half of the 3rd floor, for example. I happen to be in the sales department myself, and there's a piece of software installed on every desktop that hibernates the computer at 20:30 EST (with a half-hour countdown to that point). My department shuts down at 19:00, no other sales department is open past 20:00. We all open at 08:00 the next day, and the automatic hibernation sets an alarm to wake up the computer at 07:45. Alternately, if you turn your own system off through the start button and shut down, it'll stay off until you turn it back on.

We've also got computer recycling programs in place, and the "plant a tree" initiative where you can have us plant a tree for every computer you buy.

Sure. Some companies don't take going green seriously. But some do. And the number of companies that are taking it seriously is growing. Besides which, every little bit helps. Do you know the amount of energy that could be saved if everybody unplugged those electronic devices that "sleep" when they're not being used? 2W doesn't sound like much, until you multiply it by half a billion devices.