Slashdot Mirror
When Does Powering Down Servers Make Sense?
snydeq writes "Powering down servers to conserve energy is a controversial practice that, if undertaken wisely, could greatly benefit IT in its quest to rein in energy costs in the datacenter. Though power cycling's long-term effects on server hardware may be mythical, its effects on IT and business operations are certainly real and often detrimental. Yet, development, staging, batch processing, failover — several server environments seem like prime candidates for routine power cycling to reduce datacenter energy consumption. Under what conditions and in what environments does powering down servers seem to make the most economic and operational sense, and what tips do folks have to offer to those considering making use of the practice?"
Like when someone posts your domain name on slashdot!
You can't take down a server that's already off-line.
I have always been afraid to shutdown the servers whenever it is not completely necessary. There are other ways to cut down on costs for most setups. I can't really think of a setup or situation where I would power down servers to save money.
you see the Windows logo appear? (sorry, couldn't resist)
Attention all planets of the Solar Federation! We have assumed control! - Neil Peart
The best time to shut down the servers is right before you quit your job. Password-protecting the BIOS first adds value too.
Put redundant/failover servers into a sleep state and enable WOL.
It's pretty much up to your business....what must run 24/7, what systems are likely to get accessed in off hours, and how likely is that, and how critical are they? With redundant systems, can there be any downtime while they are powered up, or should it be immediate failover? If you use virtualization the redundancy should be easier to manage in many cases...you may be able to immediately offload to running systems and power up backup systems and then bring the VMs up there.
It's hard to get very specific without knowing your business and what you are running and what the needs are.
As our way of thanking you for your positive contributions to Slashdot, you are eligible to disable Slashdot 2.0.
I'm glad this was posted to "Ask Slashdot" where your audience is highly seasoned professionals that can give you wise, insightful answers...
In the data center that I manage, I use a few simple rules to determine when I power them down.
1) If the server is on fire
2) If there are no users using the server
or
3) If the power company is saying that I haven't paid my bill and they are sending "Hank" over to cut me off
4) Civil unrest, tornado, earthquake, zombies, etc.
If you virtualized your servers, you could create a managed power-down/power-up scenario. In the morning, your servers would turn on, your virtualized instances would move around (so they have more power for the day's activities), and then at night they'd retreat to a smaller group of servers. The unused servers could shut down for the night. You could even rotate which servers stay on overnight keeping the virtual servers running to spread the wear around if there is some.
I thought we had this discussion last week.
Does it make much of a difference if all your servers plug into a rack mount UPS that draws the same amount of power regardless of the devices running?
The greatest revenge in life is massive success.
Seriously, this analogy makes sense. When a car is cruising on the high way it's able to maintain speed using 4/8 cylinders. Servers could be cycled in a similar fashion. Do you really need 20 web servers running at 3 am on a Sunday?
.. your business doesn't depend on it.
Seriously .. powering down failover boxes or something like that is not wise thing to do.
Imagine in some fucked up situation, when your main systems goes down... you can't boot failover servers for some reason ... long fsck, or whatever.
You can power off the servers that aren't critical .. Why question on slashdot for that ?
Logic anyone ?
Makes sense to me, if your business hours of operation are 7am-6pm, to leave the servers to do their regular daily maintenance from 7pm-9pm, Monday-Saturday. Then, go to a power-saving status from 9pm-5am, rinse-repeat daily. Heck, even leave this for a 7-day schedule so you can do patches and other non-automated maintenance on Sunday.
I know it's the accepted belief in the industry that rebooting servers/downtime can be bad to a server, but couldn't you spin that the other way around and say that running a server 24/7 is bad too? Couldn't there also be benefits from letting a server rest as well?
How many boards would the Mongols hoard if the Mongol hordes got bored?
Obviously, the best time to shut them down is when they are running a Windows Server OS of any type. Also it saves you money on chairs and sheetrock repair fees.
Tech so the system can drop to a low power mode.
Also get rid of the AC to DC and then back AC then back to DC part and only have 1 AC to DC step.
When you're sure you don't need it to come back up.
You don't make the poor richer by making the rich poorer. - Winston Churchill
How many of us have servers that don't need to be live? Yeah, I guess there might be a development server, but that assumes that you're not developing. There could be a failover server that does nothing when the primary hasn't failed, but in that case you'd want to be damn sure that the failover will come online without difficulty when it needs to.
It seems to me like it would be a pretty rare case when this is applicable. I'd sooner be interested in asking, can they build servers that can selectively power down subsystems that aren't currently in use, sufficiently enough that there's no serious harm. For example, I'd consider putting some of my fileservers' hard drives to sleep over night, but I'd still want the server to be available and the drives to spin back up if I log in from home and need access.
Mostly, I'd say that if you have servers that you don't need to be live, you might not be using your servers efficiently. It may be worth looking into setting up some kind of VM server with various images that can be brought up on command. But hey, if you do have a server that you can turn off without causing problems, go for it.
Why have 16 terminal servers (sorry, couldn't think of anything else) running when no more than 10-20 users are on it after working hours? Then in the morning, power them back on again using WakeOnLan.
And that backup server with a whole lot of disks? Why not only have it running during the night when stuff is being backed up?
In the last two office moves, I lost three old servers (10+ years). I have a gut feeling that this was due to powering them down, then up. They had been up for several years without any problems. I moved them myself so I know they didn't get banged up.
Long story short, I have servers that I am afraid to power down (but too lazy to do anything but make sure they are backed up regularly)
Er... yeah, let's power down our backup servers that are there as a safety net. What could possibly go wrong?
I guess these guys don't care about little things like uptime, then?
"16MB (fuck off, MiB fascists)" - The Mighty Buzzard
Powering down your servers tends to introduce response issues. :-)
Some servers, like the HP ProLiant line, has power management features. Try experimenting with features like these first.
"Servers that sit in idle state for long periods of time are the top candidates for powering down between uses."
Then virtualize it or combine its function with another server. I see this part of the article as a bad example. It starts by saying that virtualization has helped, and then uses an example that virtualization would solve, NOT power-cycling.
Maybe its just me, but when I think of a server, I think of something important that is running, that needs to be accessible on something other than a glorified desktop. If it is important, then it cannot be turned off.
Bearded Dragon
question: when does sleeping makes sense for you?
answer: when you're tired, not busy, and a short nap wouldn't help.
In other words, if in your datacenter the servers load goes from 100 to 10000 users in minutes, no matter if it's during day or night time, you simply don't put them to sleep or you would actually lose energy and horsepower, then money.
If OTOH you have predictable hours when the systems are underloaded, you could define some rules to turn on unused servers and move some load to them (connections, virtual machines, etc).
Be careful to provide some form of hysteresis or you end up with a system that spends its whole time turning on and off stuff with nasty (expensive, potentially destructive) consequencies.
Full disclosure, I work for Citrix. Check out XenServer, which can remotely provision server workloads to virtual and bare metal machines - based on load, you can remotely power up resources as needed. I have seen the future, and it is awesome. And green.
Atari, System V, C64, Amiga, College Unix (?, I wasn't like root or nothin'), Mac, WinNinetySighs, NT, Novell, Win2K, So
Umm... Whenever you want them off?
It must have been something you assimilated. . . .
My guess is that managing energy consumption by powering down servers is the wrong way round - there seems to be a fair bit of interest in developing hardware that manages it's own energy consumption without loss, either in additional power to bring it back up to speed or in processing lag, etc. Of course, this doesn't address the poster's immediate concerns to which I have little to add other than it's probably good to cost in heightened risks of hardware failure and therefore the costs of unscheduled downtime.
If you're looking to save power, try using cpufreq on Linux, or power settings in Windows Server instead. If you simply shutdown everything by building policy, then have chron or schedular sync the file systems, then do a shut down at the chosen hour, then power them up ten minutes before the start of day (unless you have backups, reports, etc. to run
If the power cost doesn't make any difference, power them down 2x per quarter to blow the dust and crap out of them. Then keep them on if you're already green. Otherwise, power cycling is somewhat traumatic and in my belief shortens the life of disk drives more than anything else, then power supplies. Just my 2c.
---- Teach Peace. It's Cheaper Than War.
Right in the middle of a user having completed all of the form and about to hit submit button. Boy, I'd like to see the face of that user!
While there may be machines which do not need to be running they should not be refered to as servers in the traditional sense.
More hardware failures occur between powering down, and finishing booting than you can shake a stick at.
1: Can your service be load balanced across several identical servers?
2: Does your services experience predictable but varying load?
3: Can the state used by your service be rapidly replicated (10 minutes) across newly booted systems?
Not all server systems make good candidates for shutdown. Web farms do tend to because they fit the criteria above.
Deleted
Try virtualizing your *nix boxen on z/VM, on a z/10 mainframe - especially if your business/organization already has a mainframe. z/Linux is just Linux, after all... Apache and Mono are already ported, among many many other things; what's not ported can be ported in the usual way. The advantage is that you can run virtual servers on the same hardware as your mainframe "legacy" apps, without drastically increasing power consumption.
The problem is the PSU, which fails most often during power-up. Leaving the servers always on has the advantage of avoiding that particular failure mode. Also, other components in the server are prone to failure during power-up, way more often than at steady state. So, powering up your computers is overall a risky moment.
"The agriculture ministry is not in charge of Gundam" - Japanese ministry official.
VMWare Virtual Infrastructure can already turn off unneeded physical servers and power them on again when the extra capacity is needed. All the VMs (your actual server that run websites etc.) keep on running, but if they aren't using much CPU/RAM then you can squash them into fewer physical servers (automagically) and power off the unneeded hosts.
"lack of quality control is one of the pillars of slashdot"
but wouldn't it make sense, if you have say, 10 servers, to keep 9 of them on and 1 off at all times, doing maintainance on the one that is off, and constantly rotating that?
Something interesting is that most PSU's do not completely depower the system due to Wake on Lan/Ring/Timer settings. Due to this, why in hell don't you configure your servers to use STR5 when they're in standby?
If you use static caching or an SSD, you can have the system picking up excess load almost instantly yet while it's sleeping, it's not using anymore power then a system that's off.
Mod me up/Mod me down: I wont frown as I've no crown
The real problem with powering down servers is that you won't know there's a problem until you power them up again. The result is that the problem always occur when you need the servers (otherwise you wouldn't be turning them on). This instead of the problem mostly occurring when the servers are not in use or at least not all servers at the same time.
If you power up 1000 servers in approximately 15 minutes (once per day) and 10 don't power up, then you have 10 problems to solve asap. If you don't power up 1000 servers but they also fail approximately 10 per day then you don't have 10 non-working servers at the same time, but randomly distributed over the day. Meaning the problems don't queue up.
VMWare has some cool functionality such that if you virtualize all your machines, at night time when the loads are lower, you can consolidate all your VMs onto a smaller number of physical machines, and automatically turn off the physical machines. Then, in the morning, as the loads increase, you can automatically power on the physical machines and move the VMs back onto these physical servers to handle the load. Not sure what it's called but when I heard about it, I thought it was really cool.
It seems to me this is available in *some* hardware. Maybe the parent wants to check into that?
I've got racks of Compaq DL380's and I don't even know if they have Wake on LAN, but they've got other features that do the same thing.
In my environment, service response time is the primary performance metric, saving money by limiting power consumption is viewed as counter productive.
http://www.maxineudall.com/2010/02/should-economists-be-sued-for-malpractice.html
As long as power use is built into the fixed price I pay for the cabinet I rent at the colo, I'll never turn off my servers if I don't need to. Why would I?
but for something to spend 10% of its time under maintainance probably means you're doing it wrong
and surely its better to find a save money by reducing cooling costs rather than shutting down servers, or if youve got spare capacity why not rent it out to someone who doesnt what to have to look after a server themselves to justify it being on
Once servers are virtualized it becomes trivial to run only the number of virtual servers necessary to handle the load. Cloud computing, in essence. A truly distributed model of computing would work just as well, but my guess is that will only arrive sometime after most servers are already virtualized.
The only impact shutdown and startup should have is on hard disks; all other electronics should take millions of power cycles without any problems as long as the power supplies are gentle. Hard disks for virtual servers would generally live in a SAN anyway, and powering down sections of the SAN would be possible. Some drives already support a low RPM standby mode that lowers power consumption without the danger of cold startups that wear out the spindle motor. And really, shutting hard drives down completely at night and starting them up in the morning would put less load on them than most laptop drives, which already last a couple years at least in a much worse physical environment.
Powering down non-critical servers make sense:
If a large and potentially long lasting Hydro power outage occurs, and your Data Centre UPS switched over to using your generator, and you want to conserve Diesel / length of uptime your generator can keep the DC up for.
Adeptus
No trees were killed in the making of this post; however, many trillions of electrons were horribly inconvenienced.
...just how much pr0n gets done at 3a.m. on Sunday? Really, man!
There exists no way of exchanging information without making judgments. --Bene Gesserit Axiom
This question is a bit of a non-starter. If you can power it down, and no one screams, then it really isn't a 'server' at all. A server serves things and brings with it a promise of availability. If you're not providing the availability, then you really only talking about some other kind of computer, not a 'server' at all.
With the countless geeks with their *nix boxen running 24-7 at home for their unabashed uptime competition with themselves - I think datacenters are pretty efficient considering their purpose and great lengths they go for critical support.
I think there's bigger fish to fry, like the great idea of changing daylight savings time, which in the end didn't save us any energy but sure as hell cost the IT industry a pocket's worth!
P.s.: I turned off my spare 24" Dell display just for a good conscience for the last 3 minutes, there I did my part for the year....
We recently had to power down our main data centre, three times in fairly quick succession due to major power work that had to be carried out (no building UPS or generator unfortunately... boo!). Doing it was all well and good, but so many things are inter-connected, we found we almost had circular dependencies of things, so we had to be very careful in shutting down and bringing back services. The end result was something different every time wouldn't shut down, and something different every time wouldn't come back right, causing downtime to users. In testing and development it's never an issue when you're bringing something new up. It's when - six months down the line - you've integrated it into everything that you're really into a tangled web of systems, and taking it all down results in much hilarity. We were shutting down 90-odd servers; Windows, Linux, Solaris, FreeBSD, and various other things like Cisco CallManager and other random network kit. It's NOT something I'd want to do again, and having it all scripted would scare the crap out of me quite frankly. It'd require so much testing and checking, you'd be bringing things up and down all the time and causing more trouble than just keeping it up. Try and save the power elsewhere - raise the A/C temp, or virtualise some stuff.
Right, because none of this stuff can be automated...
Moron.
So many of the comments are assuming that the server's specific identity is tied to a function.
In a large VMware environment that leverages shared storage, any server can run any virtual machine.
If there are a number of standby servers that can assume additional load, the failure of one of them on startup only affects total resources available but not virtual machines or services.
We took our server estate down by a factor of 20 using modern virtualization hardware appliances and hypervisors. Thats right not 10x fewer servers, but 20x. With the right combination of investment in quality shared storage, dedicated virtualization appliances, and the skills to make them work, today our cooling, power, and rack space bills arent worth worrying about.
We are rarely at the cutting edge of technology, but if we can do it, you can.
Perhaps we could invent a way to power down servers in a manner that would not cause sudden temperature changes? What about cooling the server while it's on, then warm it while it shuts down, then let it cool gradually again, and then start warming it before we go to switch it on again, and only switch on when it is already warmed? Maybe we could think of a way to keep every chip and every component in a stable temperature and only allow very gradual temp changes. Then temperature change stress would be eliminated.
There could be some security advantages to powering off certain servers when they aren't being used. Can't hack 'em if they're off (unless you exploit Wake on Lan functionality).
But really, why buy a server if you aren't going to use it? Sounds like you need some virtualization to consolidate underused services.
You should at least consider adjusting any power saving abilities the hardware.
Power off workstations, not servers.
when you've identified an issue that started with "White Rabbit Object" ...whatever it was, it did it all
Perhaps the most hilarious thing I ever found on Slashdot.. http://www.thewebsiteisdown.com/
I will not be pushed, filed, stamped, indexed, briefed, debriefed or numbered. My life is my own.
I know of a server in a local restaurant. He often takes a "power nap" just after the lunch-time rush is over. Having conserved some energy, he wakes up refreshed and is can get back into high-power mode for the evening meals.
Why are you running an entire server, when our datacenter (and I personally admin over 300 servers at this location) runs at an average of 30% efficiency?
We are migrating (on mine and one other SA's recommendation) to a totally virtualized environment for development and testing. Production is still a no-no, but if we can show the gains with virtualizing the first two, it's easy to make a case for the last.
We can run a single blade at close to 80 or 90% efficiency, because we're using cores individually, memory, etc... and VMWare is getting pretty smart about it. I am still investigating Hyper-V, and while much of the /. crowd here would tell me not to because it's an MS product -- most of our servers are Windows based, and Hyper-V is dirt cheap compared to VMWare.
That said... virtualization is the answer, not power down servers. You get multiple benefits, and save in overall costs with power, hardware, cooling, and utilization.
The price is always right if someone else is paying.
The right answer to reduce power consumption is to make better use of the CPU cycles. Today we waste an insane amount of power because of badly optimized code. We have to much redundancy because of bad code quality. This is the area where its possible to save most power, especially in smaller enviroments .
Virtualization do not save power in all cases. Often its the other way round because most virtualizations snag a fair bit of performance. Small underutilized servers fits but any utilized server will draw more power virtualized than by its own.
The strain of heating and cooling will bring most systems down much faster than if they are constantly at the same temp.
HTTP/1.1 400
so powering down servers might not make sense for every IT department.
If a company only has operations from 9 to 5, then it would make sense to power down the servers after everyone else has left and the Admin is the last to leave and powers down the server. Then the Admin comes in early or another Admin and powers back on the server before 9am.
For a 24 hour IT department it makes sense to use low power settings to shut off hardware when not in use and use power saving screen savers, and sacrifice some CPU speed for power saving features to save on electricity.
Just remember that some companies have help desks that operate 24 hours a day and need server access to resolve issues. Either that or they have three or four shifts for employees to work with and need server access.
Remember, Slashdot does not have a -1 disagree moderation, and no, troll, flamebait, and overrated are not substitutes.
What I'm basically asking is; what about a battery solution that would charge during "off-peak" hours and then you could run your server(s) off of battery power during "peak" hours? I've heard of this being done in some smaller data centers and they've had some pretty good savings on it.
But what kind of solutions are out there to do this? Anyone got a few good links?
No really. On certain times of the year (ie. not summer), I reduce/eliminate the cooling of the server room AC and redirect the server waste heat to warm the rest of the office. Ambient air from the office space is ducted/filtered in near the floor, and a 300 CFM fan takes in the heat at the ceiling above the server tower. I estimate I capture between 9,000 & 12,000 BTU of heat an hour because of this; greatly reducing the HVAC needs of the building during the night.
This was a large part as to why the EPA gave my company one of their Annual ENERGYSTAR Energy Conservation Awards.
It never made sense to me to run an AC unit when it's snowing outside.
http://www.energystar.gov/index.cfm?c=sb_success.sb_successstories2008_johnsonbraund
I would think that if any servers can be regularly powered off, then they probably fall into one of two categories, they aren't really needed or they are only loaded at certain times. In both cases there's a good reason for consolidation, whether it be physical or virtual. obviously virtual gives the best bang for buck. I run several ESX clusters and despite dept's not trusting virtual servers, they all come around in the end. I think virtuals really are the only way to go to really save money, space, power, and the all important UPS load. Don't forget, the myth about hard disks dying after they have been running for a long time and then allowed to cool after a power failure, well that myth isn't a myth, it happens, and it happens a lot.
I cannot think of situation where powering off a server that is needed provides any benefit whatsoever, you might save a few dollars, but as we all know when you're IT dept struggles to get budget for anything, the risk of failure and the HUGE costs associated with that far outweigh ANY power savings you might achieve.
I don't give a rat's arse if powering off a few minor servers saves $100 a month, when if the disk dies after it cools down, and I then have to go into repair mode to find out what peculiar apps were installed on the server, somehow scrounge another system with no budget, and then rebuild the whole damn thing, only to find out the developers have changed so many configurations it takes weeks or months before everything is really working as it should. Even more joy is that usually occurs at night if power is lost, as we have nothing better to do anyway. NO THANKS!
For companies who have massive budgets for IT, and routinely swap out hard disks, etc, maybe that would work, most companies I know and have worked for over the years though tend to view IT as a parasitic loss centre run by people who spend there days watching the blinkety lights and having fun, "it's just a bunch of servers, how hard could it be" or "how much money could they need, I can go the PC shop and get a whole quad core rig for $600!"
But I will say, nothing gets Capital Exp forms signed faster than a major downtime!
... is with electronic voting servers, to force paper ballots and more accurate counting.
Have a single server that handles all fail-over needs, and is constantly running. Isolate this server from direct end-user interaction as best you can.
In the event of failure of anything, switch to your single fail-over server. Bring up, populate (if necessary), and then switch to a second, dedicated server from there.
You have a few database servers, several Web servers, FTP, e-mail servers, etc, all isolated and serving different purposes. Each of the critical ones-- notably every database server, the e-mail server, the customer-facing Web server, and any Web servers hosting mission-critical applications-- has its own fail over. Why run two dozen extra fail-over servers when you can run one or two (isolate the out-facing Web site ones from the internal, critical function ones), and use them to keep functionality going for the 10 minutes it takes to bring up the dedicated fail-over box?
Support my political activism on Patreon.
Doesn't anyone do backups at night? When I was a sysadmin, I learned that backups during the day = bad, so if you wanted backups done, they had to be done at night. Can't do that if they're powered off. I suppose WOL would help reduce the time they're on...
This is exactly what my blade chassis does. I get a power savings of about 1/3 over 1U servers. And it's just a low end Dell chassis - nothing special.
Do not fold, spindle or mutilate.
Virtualization is really your best option here as if you have specific purpose boxen then your going to be to reliant on them actually restarting.
You see on the inside of most of those boxes with the flashing lights there are many moving parts...spinning hard disks that rotate at 7000+ RPM, Many fans, 1 for the PSU 1 for each CPU, Generally 1 or 2 for air circulation also.
Now as with most moving parts it takes more energy to begin the motion than to maintain it which is why you have more load on the Power supplies than when your actually just idle. As pointed out on other comments you will more often see Powersupplies die on power up when the capacitors are being charged and the load is spiking.
the second concern is that the motors in Hard Disks and fans create heat which causes things to expand. When they cool the metals etc contract and you end up with a drive that may or may not find it easy to restart. Obciously hard disks are always being developed but unless it's solid state you will have a generally expected lifetime....MTBF is the manufacturer term Mean time before failure. In general constant rebooting of systems drastically reduce these values. Dust also settles if there is no airflow from fans also possibly creating issues.
If your virtualized you can move processes to and from boxes that fire up and shutdown without losing functionality just performance.
Google for instance can basically just plug things into their network and the grid starts to use it. If something is dead then theother 9 million PCs do its job and a flashing light appears for some guy with a trolly to remove them and put a newer faster prettier box in its place.
You can also do little things like power down monitors, Spin Down hard Disks, Speed Step CPUs etc which would also provide some options also....
most Mobos nowadays have "some" reasonable power saving options
If you run Coreboot as BIOS, then you can boot your servers in two seconds.
So then it not so much of a pain to power down the servers.
If you run load-balancing, and its at a time where its low load, then you can power down some of the servers.
http://www.coreboot.org/
http://en.wikipedia.org/wiki/Coreboot
Basically what I'm asking is what about charging batteries during "off-peak" hours and then running the server off of the batteries during "peak" hours. Add to that, I don't know of any product out there that's offering this. Does anyone else? With the cost of batteries, this might be a non-starter, but just asking just in case.
If the company uses time of day electricity pricing from the utility, electricity is very cheap overnight. It would in such a case make sense for a company to leave the servers on 24/7 and try and run as many batch jobs as possible overnight.
I used to believe this. I have looked at data in my environment. My project reboots servers by cutting the power supply at a managed power strip. Power is brought back to the system by re-energizing the power strip port.
My findings after 5 years of doing this (which many machines experienceing this multiple times per day), no noticable rate of power supply failures over a large installed base (less than 1% of all failures). These are cheap power supplies at that.
Point is, machines are much more capable of surviving the "shock" of a power up/down sequence than your instincts would lead you to believe
I love all this green crap.
right now im scheduling tasks to wake up 400 desktop computers nightly at 3am then off again at 6am to do wsus updates and a defrag
Hell, i should make them fold too just to guzzle up some more power!
Environments where it doesn't make sense are any where you have only 1+1 redundancy, or you can not effectively load balance live load.
To make this all transparent it is important that your environments require at least three or more servers during peak loads, so when load falls you can power off servers and still maintain some level of redundancy. If your environment has 10 identical servers, and one doesn't come back in the morning, it isn't going to cripple you to run on 9 while you get the vendor in to replace the failed one.
To be most effective it required true application-level clusters where workloads can be moved through load balancing. It is no good when at the end of the day you have 10 terminal servers with one or two idle sessions each - you can't consolidate the workload back to two servers and switch the other eight off.
VMware is good, but you need to either power off the VMs at night, or have a few nodes with lots and lots of memory to hold the RAM requirements of the mostly idle VMs
In the good old days thermal issues used to be a problem - socket mounted ICs would 'walk' out of sockets over hot/cold cycles. Now all the chips are hard soldered. Now we have redundant PSUs in most servers it doesn't really matter so much if a PSU fails... just pull it and install a new one!
At the opposite end of the spectrum I have my own Linux based server that powers on when a PC boots (using a magic packet utility in the startup) and powers itself off after if nobody is on the network. Works well for me.
large datacenters are begining to use power monitoring systems and branch circuit monitors.. such as FieldView... http://www.energy-options.com/product_fieldview.aspx to manage the power in the datacenter, rather than shutting/powering down servers - they better utilize the power within the datacenter so shutting down/powering down servers becomes obsolete.
I work for a medium sized company with hundreds of servers in our main server room. (I'm not including the 100's of others in smaller labs, etc.)
Its a huge undertaking to power down servers. There are some pretty complex relationships between applications and servers, so just powering them off isn't a simple task.
Every few years we have to do a complete shutdown for electrical work. Guaranteed a few servers will have problems and not power up. So, given the man hours + cost of failure I can't see how in the end we would be saving any money.
I'm sure there are a few servers that could be shutdown but at the end of the day I don't think we'd be saving much on power.
A few years ago we consolidated environments on newer hardware and got rid of the old. We could support more environments on less hardware, so we saved power costs but more importantly licensing costs. Just by consolidating a few Oracle servers saved us a boat load. I would say that approach would save more money than shutting down servers when idle. So if you have idle servers, consolidate.
"Thanks to the remote control I have the attention span of a gerbil."
Just don't power them off.
What, your CPU can't reduce its clock speed when idle? Stop using fucking x86 then.
Honestly, unless your company has less than 10 servers, I can't see a scenario in which you want to do this.
Increasingly hardware manufacturers are coming under fire about their green credentials, so it will be no long before this becomes a non issue.
IANAL but write like a drunk one.
Shutting down seems wasteful. I would:
1) If you're shutting down outside of nighttime, you're wasting resources and can consolidate by virtualizing on fewer hosts
2) Enable power management on your servers. Things like SpeedStep and HDD power management (or even spinning down certain disks) can go a long way to reducing power requirements when load is light
3) Suspend, don't power off; you can configure your servers to go to sleep after not receiving any traffic for a certain period of time, and then automatically wake up and handle the request when they receive ethernet traffic. This might add a few seconds of delay to the first request when waking up, but if it's the first request of the morning after being asleep for the night, that's probably not a big concern.
Combining these three could produce significant power savings without the downside of shutting them off; powered-down servers can't answer requests at all.
I always make it a habit to power down my server after work, since it's not allowed to be used outside office hour.
If you delay pleasure infinitely, the pleasure will be infinite. (YM)
I worked for a client that used a farm of web servers tied to their multiple Oracle systems. The web servers were all Sun Ultras balanced using Resonate as the balancing agent.
During prime time, we needed over 60 servers running but, between 6PM and 6AM, we only required about 15 to handle the load. By taking 75% off line every night (not always the same 75%) we reduced power consumption a great deal. By also shutting down 4 of the 6 Sun 6500s, we also reduced power of the data center.
In a year's time, we conserved over $80,000 in power alone and, had plenty of opportunity to perform off-hour upgrades and maintenance.
Failure rate due to power cycling was immeasurable.
Banjo - The more I know about Windoze, the more I love *nix
Our backup servers power up nightly, rsync to the live servers, then shut themselves down when the rsync is done. This saves electricity for the server and the AC unit. It is easy to do with BIOS wake on RTC.
It never makes sense to power down a server that supplies web pages, if you are in e-commerce.
The moment your potential customer sees a 404 page, you've lost him.
Indeed we design all our pages to be displayed in less than 7 seconds, as our research showed anyone typically waiting longer than that for say "search results" would be likely to go elsewhere rather than waiting.
"Full disclosure, I work for Citrix. Check out XenServer, which can remotely provision server workloads to virtual and bare metal machines - based on load, you can remotely power up resources as needed. I have seen the future, and it is awesome. And green."
I'd be more trusting of your employer's ability to code something like that if you could provide me with something simpler - like a Citrix client that can handle two monitors with different resolutions correctly, and that doesn't get confused from time to time about where my cursor is.
That bondi blue iMac is so last-century, I wouldn't even put it in my ass. Kinda like your joke really, stop eating leftovers out of your belly-button and check out the real world sometime, a lot has changed since 1997.
...you don't have power.