Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[blgentry]

My workstation based MC Client starts with a local Library because I often do testing on local Libraries, and I don't want to wait for the Server to wake and the Client to connect on every MC restart. (The MC Server often goes to sleep while I am playing with stuff, so it needs to be woken most times when I start the Client.)

Servers should never sleep.  That's the point of being a server:  To serve.  Letting them sleep or turning them off does nothing but cause problems like the one you just described.  I don't mean to sound accusatory.  I just grow weary of all of these problems that are caused *explicitly* by people turning machines off or letting them sleep.  Computers work better when you leave them on all the time.

Brian.

[JimH]

Servers should never sleep.  That's the point of being a server:  To serve.  Letting them sleep or turning them off does nothing but cause problems like the one you just described.  I don't mean to sound accusatory.  I just grow weary of all of these problems that are caused *explicitly* by people turning machines off or letting them sleep.  Computers work better when you leave them on all the time.

That's your opinion.  It isn't a fact.  There are servers and there are servers.  Some must be up 24x7.  Some don't have to be.

I have two machines I use as servers and I'm not very careful about either.  One sleeps every night.  I only wake it when I need it.  It's trivial to do so, and I don't have problems with it.

Putting that server to sleep saves about 40 watts an hour, or around 800 watts a day.  Over a month, it's about 24 KW hours.  I pay around 12 cents a KW.  It's about $35/year.  I'm also the guy who goes around turning off lights.

Windows is pretty good at this now.  It wasn't at one time, but sleep works now, and it has worked for a long time.

The other server is an IdPi and it never sleeps.  Once in a while, I yank the power.   

[blgentry]

That's your opinion.  It isn't a fact.

No, it's a fact.    Seriously, it's a fact.  I maintain a rather large infrastructure of computers as my day job.  I've been doing this for over 20 years.  I have some data about this.  Here's an anecdote for you:

Hurricane Irma recently swept through my city.  It was "only" a tropical storm when it hit us, but it took out power in a lot of the city.  As a pre-emptive measure, we shut down one of our Data Centers entirely.  About 350 computers, plus one major storage subsystem.   Days later, when we turned everything back on, we mostly had no issues.  Except we had 6 or 7 computers that wouldn't boot.  Fiddling with most of them got them back up and running.

The storage subsystem had a software problem, which was just that:  Software.  Had we not turned it off, it wouldn't have happened.  So this was a software bug that was triggered by turning the systems off.  A direct consequence of not leaving the system running.

Then the drives started failing.  I think we lost a total of 5 drives over the next 5 days.  Not a ton of drives for a system this size.  But it's about 10x our normal drive failure rate.  Caused by turning the systems off and back on.  There's no denying that.  We never lose 5 drives in a week.  But we did this time because of the stress induced by the off and on cycle.

You can cite your power savings all you want.  If that makes you happy to be saving energy and money, more power to you.  But it's not good computing practice and it absolutely, positively shortens the life of some of the components.  You can choose not to believe me.  I trust my experience and the experience of those I work with.  Leaving computers on makes them run better.  I would invite you to cite any reference you have for your belief, as I have mine.

Brian.

[Hendrik]

If your systems don't cleanly reboot when they need to, then your systems should really get better maintenance though. It can always happen, even if you don't want it to. Even more so if your hardware dies from a simple power cycle, it was probably on the brink of death already anyway.

[RD James]

Most home servers don't need to be on 24/7, and I'd argue that it may be preferable to shut them down or reboot them more frequently if they don't have ECC memory.
I do keep my main server running 24/7, hooked up to a UPS, and don't allow the hard drives to sleep, as drives do survive much longer when you do that in my experience.
I've also thought about setting up fan curves to try and keep things as thermally stable as possible instead of always running at a high RPM and keeping everything as cool as possible, but the tools for that aren't great and constantly fluctuating fan speeds can be annoying.

[blgentry]

Here are the key points for me:

A.  If you leave servers running, the services are ready to be used any time.  This eliminates all of the problems people write here about on a weekly basis, about things that don't work because servers are off or asleep, drives are asleep, etc.  No one can argue with this.  When servers are on, and not asleep, they are ready to do their jobs.  Simple.

B.  Hardware of all types fails faster when you power cycle it.  Incandescent light bulbs almost always fail when they are turned ON.  It's rather rare for a light bulb to die in operation.  It's usually the on/off cycles that kill them.  Computer hardware is similar.  If you don't believe me I guess I won't argue.  Instead I'll ask, how many computer systems have you maintained in your life?  I've worked with a fairly large number.

Today's obsession with lowering power consumption runs directly counter to computer system reliability and lifetime.  When you decide to make everything turn off or go to sleep, you necessarily are making things harder on yourself.  As long as you realize that, you'll make an informed decision.

Brian.

[DocLotus]

I'm in both camps.

I was a consultant / programmer for 18 years before retiring. I maintained several small customer systems over that time & noticed two things...

• The customer would never turn their individual personal computers off. After a week (or more) of on time I would get a call saying something like "My computer is messing up". I would go to the user & ask "How long has the computer has been on?" "Oh, three weeks or so." I would simply reboot the computer, advise the user to do the same at least once a week... problem resolved (until they forget to reboot occasionally). Seems that Windows sometimes messes up after running without a restart for days on end. I believe it has to do with memory getting messed up as sometimes a simple shut down & boot up won't clear the problem but a complete power off (throw the power switch or pull the power cord) will resolve the problem.
• Computer has been running well. Shut down the computer at end of day with no problems. Next morning boot the computer & Windows messes up... won't boot, flaky boot, problems after boot up, etc. Seems like Windows LOVES to mess up sometimes on boot after running just fine for days on end.

So, as you can see I understand both sides of the argument.

Cheers;
Doc

[ugee]

Edit note: removed irrelevant remark(s).

[blgentry]

Recommending to keep things running and no problem will occur is still a problematic advice, and at times not even a practical one. You've told an anecdote about that - if your assertion would be entirely true the right thing to do would have been to run the DC from backup power for days /weeks / ... My guess: that was not feasible or at least from an economical point of view not very reasonable. 

What's ironic about my story is that our DC did *not* lose power during the hurricane.  We turned everything off because the person in charge was paranoid based on a previous event that caused some bad things to happen.  Had we kept everything running, we would not have had the various failures that I told you about.  So there was no need to run on backup power.  Even if there had been, it likely would have lasted less than 48 hours. 

But yes, we temper these decisions with how important it is to keep each type of computer running.  In many cases we can easily make the decision that there's no reason to run these computers and just shut them down, should there be a long term power outage.

At a data center level, if you start analyzing power consumption, it will make you a little nuts.  We have individual racks in one of our data centers that can pull 30 kiloWatts when all the nodes are doing a lot of work.  ...and they do some serious work with all that power.

But we are getting way away from the point.  I'm just answering some of your charges.

Brian.

[tzr916]

I do not let my MC server sleep, and I have 4 client PC's that I don't let sleep either. I figure I'm saving more than enough money on cable company box rental fees to justify leaving all my stuff on 24/7. And I simply don't want to gamble on having any possible sleep related issues. Basically only reboot a PC when something goes wrong or if there's been a Windows update.

[JimH]

It's not just about money.  Global warming, too.

[DJLegba]

The manufacturing process requires a lot of power, as does distribution. Leaving your equipment on 24/7 may increase your power consumption but still result in a net decrease of power consumption if the hardware ends up lasting a lot longer.

Using your finger to flip a light switch uses a lot less electricity than firing up an app that relays commands through your router and wireless control device. Maybe your new name for Engen could be Global Warmer.

[RoderickGI]

Well this got a little exciting didn't it! 

Basically, Data Centers running primarily UNIX platforms and serving multiple users over wide geographic locations, and hence time zones, should almost definitely always stay on 24/7. The aim here is to maximize service and hardware life while minimizing cost after all.

Home servers, and specifically Windows based consumer software home servers can reliably be allowed to sleep and will wake when required. It just takes a little time. Hence my original quoted comment. The aim here is to have a pleasant home environment and convenience while compromising on cost and hardware life.

If I am sitting at home quietly reading in the lounge room and the HTPC is on with disks spinning but nothing playing I can hear it, even with my tinnitus and strange old hearing profile. I don't want to hear it. So I let my HTPC sleep. I have been doing it for years and years with my PCs. I get more problems from bad software, dodgy BIOS and memory, chips, firmware and so on, than I do from a PC being allowed to sleep and being woken as required.


As always, different rules for different situations and requirements. 

[RD James]

The manufacturing process requires a lot of power, as does distribution. Leaving your equipment on 24/7 may increase your power consumption but still result in a net decrease of power consumption if the hardware ends up lasting a lot longer.

It's a big problem that people are replacing functional old hardware with newer more efficient hardware, which ends up creating more waste than using it until it dies and recycling it.
Or they are buying hardware for automation which has to run 24/7 and has a higher standby power consumption than they save. All those connected LEDs, power strips etc add up.
On top of that, the amount of money saved saved by sleeping or shutting down the server is offset by the costs of replacing dead drives or other hardware.
 

It's not just about money.  Global warming, too.

Depending on where you live, green energy sources may be an option. My energy provider uses 100% renewable energy sources - mainly hydro and wind. No coal, gas, nuclear, or other fuels. It costs a little extra, but I feel that it's worth it.

[JimH]

Drives don't die because computers sleep.  Old wive's tail.

[RD James]

Drives don't die because computers sleep.  Old wive's tail.

I was having to replace drives at a much higher rate before I set them to run 24/7 and never sleep.
The failure rates for WD Green drives were significantly higher than anything else they produced at the same time due to their aggressive head parking and sleeping when idle. They killed off the Green line of drives they had such a bad reputation.
If a drive costs $300+ and it's saving me $28/year to sleep the server (since it's on for at least 8 hours a day), the math doesn't work out. It's cheaper to keep the server running.

[blgentry]

Drives don't die because computers sleep.  Old wive's tail.

So you don't believe my story?

Brian.

[astromo]

All useful discussion. One of the NAS units I've got downstairs is costing me somewhere between AUD $50 - $100 to run in electrical power if it runs continuously. The to and fro here is prompting me to consider the value equation. It could also tip me over into going for solar with a battery.

So you don't believe my story?

Brian.

https://en.wikibooks.org/wiki/Minimizing_Hard_Disk_Drive_Failure_and_Data_Loss/Stress_Control

I reckon it's plausible:

Power cycling control

Shutting down and rebooting a computer or resuming it from hibernation cycles the power to the drives in the computer. The spin-up operation performed by a drive after a power cycle is believed to place more stress on the drive than running the drive continuously for a long period of time.

Based on professional experience of system administrators, it is believed that there is a direct relationship between the number of power cycles of a computer and the probability of failure of its drives. In other words, a computer with a high uptime may have a lower probability of drive failure than one that has been power cycled routinely.

[tzr916]

Been working five years in IT service for a large financial planning institution with over 100 offices just in my service area, and thousands in USA/Canada. Each office having 2 to 5 PC's that are mandated to run 24/7. I cannot count how many times I have had to replace a CPU tower just from powering down the PC for a peripheral swap/other services, then when powering on it hangs or beeps or reboots endlessly etc. Whether it's a fan or power supply or hard disk etc. The service company I work for has been doing this for 25 years and obviously knows the benefits of leaving a PC on 24/7.

[dtc]

I started managing my first computer systems in 1976.  At the time, I had long discussions with our service provider about running systems 24 hours or day or shutting them down at night when they were not being used. The real concern was the disk drives. You know, the big 60 MB ones with 10 platter removal disk packs. They looked like washing machines and big data centers looked like laundromats in those days. We tried it both ways and never came to a definitive answer, though we generally left then on 24/7.

Good to know that that discussion is still going on.

[RoderickGI]

I think there is ample evidence and no doubt that power cycling hard drives, particularly older hard drives, will shorten their life, and they typically fail when rebooting a PC. Newer drive designs are better, but they will still fail.

But for a home computer, unless you have a very large UPS and/or solar power with battery or alternate backup power supply, what will happen when you get a power failure if you have been running a very old drive that is way past reliable restarting age?

You are going to get a failed drive straight after the power failure.

We are talking about Windows PCs here. They occasionally need to be rebooted for Windows updates, depending on what you were doing at the update time and the type of update. They occasionally just need a reboot because some software mucked up, and memory usage is a mess. Or a driver locks up a process or something, so a reboot is the only way to get the PC working again.

I'd rather have a PC that I know will reliably reboot when required, until a disk fails. I cover that issue by running Hard Disk Sentinel, which I'm hoping will predict imminent failure so I can replace a drive before it fails. So far, so good.

[RD James]

I think there is ample evidence and no doubt that power cycling hard drives, particularly older hard drives, will shorten their life, and they typically fail when rebooting a PC. Newer drive designs are better, but they will still fail.
But for a home computer, unless you have a very large UPS and/or solar power with battery or alternate backup power supply, what will happen when you get a power failure if you have been running a very old drive that is way past reliable restarting age?
You are going to get a failed drive straight after the power failure.
We are talking about Windows PCs here. They occasionally need to be rebooted for Windows updates, depending on what you were doing at the update time and the type of update. They occasionally just need a reboot because some software mucked up, and memory usage is a mess. Or a driver locks up a process or something, so a reboot is the only way to get the PC working again.

While a home user running an OS like Windows may not be able to avoid having to ever reboot the system, you can still minimize the number of power cycles that a drive goes through by keeping it running 24/7 and preventing it from sleeping.
You may still have to reboot once or twice a month for updates, but the drives are not going to be power cycled daily or even several times a day.
I'd think that a reboot is at least a bit less stressful than sleeping or hibernating the system overnight too. At the very least, it's going to remove some amount of thermal cycling since they aren't going to cool down to ambient temperatures. I believe that one of the surprising results from one of Google's studies years ago was that failure rates were higher when drives were cold.

Regarding the UPS: it depends what you consider "very large" but a 900W UPS is smaller than a standard mid-tower PC case - maybe 1/3 or 1/4 the size, and can keep the system running for a few hours.
I have Windows set to immediately do things like drop the CPU clockspeed, PCIe link speeds etc to the minimum as soon as it switches over to battery power to maximize runtime, and if it is going to run empty, it will at least hibernate cleanly rather than cutting off.

I'd rather have a PC that I know will reliably reboot when required, until a disk fails. I cover that issue by running Hard Disk Sentinel, which I'm hoping will predict imminent failure so I can replace a drive before it fails. So far, so good.

You might not want to change something that is working for you, but I've switched from HD Sentinel to StableBit Scanner.
I still use HDS for things like initializing and testing new drives before I put them into production, but StableBit Scanner actively runs tests on the disks it is monitoring rather than just pinging the disk every five minutes and checking for SMART errors. I believe it aims to complete a full surface test of each disk once a month.
If you use StableBit Drive Pool to pool your disks together, it can also start migrating data off at-risk drives immediately, and prevent anything new from being written to it.

[JimH]

While a home user running an OS like Windows may not be able to avoid having to ever reboot the system, you can still minimize the number of power cycles that a drive goes through by keeping it running 24/7 and preventing it from sleeping.

Any data on this problem?

[OverTheAir]

The thread title is a bit of an absolute, especially in a JRiver MC context where servers are largely if not wholly used to store media that in a home environment is unlikely to be accessed for much/most of the day for many. Audio may be one exception to this use profile but in my case my FLAC audio rips total about 150GB so I copy and sync them up across multiple PCs so they are local for playback and don't need the main servers to be on. If I did need the audio on a 24/365 server then it would be on something sipping power.

Consider my example. I have Server1 for films and Server2 for recorded TV.
Server 1 consumes 112W while idling, Server 2 uses 90W idling, both with disks active. If left on 24/365 that's 1752 kWh. Where I am with my base usage that is likely to be on Tier 2 electricity pricing at $0.276 per kWh or $483/year or almost 5x 4TB drives per year. Certainly I could spend more money and consolidate the Servers instead of utilizing old hard drives, motherboards etc but the ROI from doing that doesn't make much sense to me. BTW Tier 1 pricing is $0.20 so even if some of the usage falls into Tier 1 its not massively lower cost.

If all the drives go into standby then that would reduce the consumption by just over 40% but that seems to be against the principle outlined previously in the thread and would still burn plenty of juice.

Compare constantly-on-and-spinning to just powering up the servers when I need them. Assume 4 hours per day. Its likely less on average and sometimes only one server will be switched on if I know I'll be watching TV vs Film but for this example I'll assume always turning on both. In this case the annual cost is approximately $80/year. The balance of $400 not spent on power buys quite a bit if I need to, not to mention broader concerns on wasteful energy consumption and the additional heat dissipation in Summer.

In addition only powering the servers up when required reduces the chance of a random power cut causing a problem.

[JimH]

Nice post.

This thread is definitely partly about cost, but it's also about the problem of global warming.

[DJLegba]

Idling a NUC 24/7 for a full year uses about as much power as driving a Tesla 50km without your halo illuminated.

[OverTheAir]

... but it's also about the problem of global warming.

Oh certainly reducing our energy footprint is something to consider too.

One thing that struck me in previous comments is the expectation/actual experience of HDD failures. I've been building HTPC's, recording TV and storing media on PCs since 2006 and in my view the only HDD failures are down to my own errors. I'm not saying others haven't had failures (or that my example is statistically valid in any way) but I do wonder how many HDDs failed due to inherently bad design of notorious drives by the respective HDD manufacturer versus any use case?

In 2006 I bought 3x 250GB drives. Two failed ... probably due to my inadvertently allowing them to get up to over 60C because I was trying to keep fan noise to a minimum. It still took years, probably half a decade or more, before they failed after that "stress test." The third still works.

A while later I bought 4x 500GB drives. They are all still working in my PCs today. One had a surge suppressor or something across the power rails fail years ago, preventing a PC boot ... cut it out with snips and the drive has worked fine since.

My most significant failures happened last year with 4x 1.5TB drives from a total of 11. These were Seagate drives, as all my drives have been, and included some notorious 1.5TB models. I was having errors in my Snapraid array. It turned out to be a PSU overload issue that I was slow to detect as the reason and which apparently bricked them when I was messing about re-building the server (although they may well still be recoverable, I just haven't got around to trying.) However, I didn't loose any TV recordings because I had dual parity Snapraid in effect. Those four drives cost ~$200 total new and it was still my fault. The remaining 7 are working fine.

And perhaps the best recommendation I can make from my non-technical non-sysadmin position for those building drive storage arrays for media storage/archiving and worried about HDD longevity/failure for any reason is to implement Snapraid http://www.snapraid.it/ and definitely not RAID 5/6, been there done that and the latter didn't work out well.

 

[thorsten]

Interesting Discussion
But I Wonder what kind of server you run: my WindowsServer2012 runs 24/7 but has an average power consumption of 13W 
Equipped with celeron P3900 and 2x4TB in raid configuration. Perhaps this is the compromise you are looking  for: efficient Hardware 

[rudyrednose]

Yup, interesting discussion.

My points:
- we are discussing home environment, not datacenter.
- there are multiple good solutions, to different needs.
- I have one server on 24/7.  It is my home file server and JR Media Server among other functions (storing music, photos, TV series and OTA TV).  It uses ECC memory.  With no RAM parity, I would not trust a Windows machine never rebooted (RAM bit rot).  Most of the time, that server only consumes 30W, measured at the plug.
- on that machine, files frequently accessed (new OTA TV recordings among them) are stored on SSD (two 1TB, plus 512GB system SSD), the rest is on 5 hard drives allowed to sleep.
- I have two other servers, Linux machines, that are files servers holding my 2D and 3D movies (large number of hard drives).  Those are always off but for when we are actually watching a movie.  They are mapped as drive letters on the main server.  I too use Snapraid dual disk parity (RAID 6) on those two machines (too much data to backup, and I have the blurays anyway), the main server do have off site backups.

I am dedicated to energy efficiency, I have an electric car and my house's HVAC is geothermal.  But I want my 24/7 efficient main server 

[OverTheAir]

Well I provided the consumption numbers so I suppose I should describe the servers for context and possible use/reference by others. As might be surmised from my continuing use of old drives in some applications, I tend to use components for as long as they are practical which will explain some of my choices. For example I still use an old circa 2002 Compaq portable running Ubuntu in the kitchen as a Squeezeplayer. My servers use Ubuntu 16.04 LTS desktop version.

So to the servers.
Server1 uses a relative recent AMD Kaveri APU based solution with 8 onboard SATA connections along with a 4xSATA port PCIe expansion board to support 10 HDDs - a mix of 3 and 4 TB drives. The PSU is 750W 80 plus Bronze so not the most efficient. I chose this CPU/board solution for several reasons.

• Its ability to support a lot or drives direct from the motherboard saved additional expense on either a 8x SATA port expansion board or an additional 4 port one
  
• The increase in processing power with an A8 quad core 65W TDP processor (over the previous board that used a single core 45TDP CPU) to accelerate the Snapraid scrub functions that perform data integrity checks on the disk array
• The ability to use it as a media player in future, if for some reason I decide to not use it for a server
• Motherboard and processor were excellent value for money

Server2 is rather less impressive as it tends to use the cast offs. It was while creating this server and consolidating a number of the 1.5 TB drives that I ran into a power limit with an old PSU that took out several drives before I realized what was happening. It is now quite stable with a 550W 80 Plus Bronze supply. It uses my oldest motherboard + CPU combo, a circa 2006 AMD Athlon 64 X2 3800+ 939 socket with MSI GeForce 6150 m/board that was my first HTPC. I expand the 4x onboard SATA with a 4x PCIe SATA expansion board to support the 7x 1.5 TB drives. While no power house, it manages the processing of Snapraid scrub operations OK and avoids more landfill. I should add that the motherboard was rescued from an early death by replacing failed electrolytic capacitors many years ago.

Its possible Server2 may be consolidated with my HTPC down the road when I move the latter out of the home theater, at which point I would upgrade the GPU to support a 4K UHD projector and perhaps upgrade the processor/motherboard for faster conversion of HDTV TS files to H.264/AVC although the advent of ATSC 3.0 may make this unnecessary.

And to get back to the original topic of HDD reliability. The HTPC I referenced has always powered up and down from sleep, often multiple times per day, because it is also my HDTV recorder and is using one of those 500GB drives from 2009 to record to. I power it up and down because it consumes ~100W due to being old (AMD Phenom II X3 720 CPU) and because it also uses a Radeon HD 5670 GPU, an AVerTVHD Duet HDTV card and ASUS Xonar DX sound card since I have no AVR.