May 1, 200818 yr OK so maybe the world knew this but was reading wikipedia and came across this citation A common misconception is that a colder hard drive will last longer than a hotter hard drive. The Google study showed the reverse -- "lower temperatures are associated with higher failure rates". Hard drives with S.M.A.R.T.-reported average temperatures below 27 C had failure rates worse than hard drives with the highest reported average temperature of 50 C, failure rates at least twice as high as the optimum S.M.A.R.T.-reported temperature range of 36 C to 47 C.[34] The paper it refers to lives here: http://labs.google.com/papers/disk_failures.pdf This is a complete shocker to me. I would have swore blind that this was not the case. There are alos a couple of follow on papers that seem to back it up. The other shocker is that SMART is just as likely to not predict a drive failure as it is. Ive totally paraphrased here so i highly recommend people draw their own conclusion by reading the source material.
May 1, 200818 yr I think the issue is more that colder hard drives with many stop and start counts are more vulnerable to failure. I've always had 2 10,000RPM drives for my OS and 4 IDE drives for my data storage. During the summer. The 10,000's ran from mid 40'sc to upper 50'sc. The 5400 & 7200 IDE drives ran from mid 30's to upper 40s. During the winter they would be around 23-30c.(My apt is around 65F in the winter). They were always spinning 24x7. The ide drives would fail around 2yrs I would get a refurbished drive and usually before failing, I would end up buying larger drive anyway. The SCSI drives failed anywhere from 1.5 to 2.5 yrs. Usually in the summer due to heat. (no a/c in my computer room). I recently put a whole new fan structure which kept them cooler (under 50c) and they lasted much much longer. With unraid the drives will be stopping and starting many times over the course of the next few years. I surmise the failure rate will actually increase. Years ago when I used to work on mainframes (with drives the size of a car trunk). Keeping the drives running all the time gave us more life, then letting them spin down on the weekends and startup on Monday mornings. So it's not necessarily the cold, it's the stops/starts when the drive is cold. I've never had a spinning IDE drive fail in the winter. even as cold as my apartment is. Drives like it in the mid60's.
May 1, 200818 yr Author OK people, let's switch off SMART *AND* our case fans! (yeah, right) Obviously thats the way forward But seriously this isnt the first time perceived common opinion has been wrong. This study far exceeds personal and collective experience invividuals have access to. i.e. Almost no people have access to this large a data set and from what i can see (only read it once) the data is quite clear... warmer drives last longer than colder drives, and excessively cold drive fail more often than excessively hot ones (they caveat this bit quite heavily in their conclusion probably because its so radical a conclusion) we find that failure prediction models based on SMART parameters alone are likely to be severely limited in their prediction accuracy, given that a large fraction of our failed drives have shown no SMART error signals whatsoever. Is quite clear and less radical since i dont think anyone would question that a drive failed without warning. But this is also interesting for unRAID users The critical threshold analysis confirms what the charts visually imply: the critical threshold for scan er- rors is one. After the first scan error, drives are 39 times more likely to fail within 60 days than drives without scan errors ... so if you see a single SMART error of the type cited then it should be assumed that the drive is basically dead.
May 1, 200818 yr what I would more easily accept is that the expansion and contraction of the plates when they get hot and cold all the time by starting and stopping, can lead to wear also a Fact of Life is that I've seen many products decay BECAUSE they are used less (stupid example that comes to mind, car electric windows... I've noticed that the ones operated less, wear out more!)
May 1, 200818 yr I wonder what would happen if we ran the same type of tests on light bulbs. Most light bulbs blow as they are being turn on. At that time their temps are low. So a high incidence of low temp failures may be more of an indication of power on failure. I have had too many heat related computer failures in my lifetime to believe that runing any electonic device hot on purpose is a good idea. I do believe that a reasonable debate on running them 24x7 vs spinning them down when not in use could probably be made. I may go into unRAID and up the spindown time to avoid having them come up a down several times a day when I am moving things around on the array.
May 2, 200818 yr I'm pretty sure I have posted my concerns about this study here - if not, it's over on avsforum. In summary: while the numbers are big, I don't find it proves much of anything. Bill
May 2, 200818 yr I think if we get a solution to the problem discussed in this thread it would go a long way towards preventing much of the unnecessary spin up wear and tear. [keeping my fingers crossed] http://lime-technology.com/forum/index.php?topic=1822.0
May 3, 200818 yr Just my own non-scientific study contradicts this. Back in the day of the Series1 TiVo, you had to completely remove the drive holder which had the fan attached to it to upgrade the drive. It's easy to forget to reattach that fan. Any time I forgot to reattach that fan, the drive would start dying. It'd start with the click of death (which would make the audio/video stutter, sometimes so bad the unit would reboot) and if I didn't yank it out right then, the drive would soon be dead (and even if I caught it early, once it started failing it was no use in a TiVo again). You want to keep your drives as cool as possible. S.M.A.R.T. isn't foolproof, but it does help let you know what's going on with your drive. If you see weird things going on with it (lots of sectors going bad all of a sudden), you can make an informed decision whether to keep using it or not.
May 4, 200818 yr I read the article a while back and took it with a grain of salt. The majority of the hard drives I've had die on me were heat related; I think only two were not. But, It does make sense that starting and stopping the disks produces more wear and tear than those kept spun up.
May 4, 200818 yr Author Playing devils advocate so i can understand better why this paper is incorrect... How do you KNOW heat caused a HDD to fail, surely in most cases it is actually "my HDD failed and it was hot therefore it broke because it was hot"? Isnt that like saying I crashed my car when it was sunny therefore the sun made me crash? How large a data set are we looking at when people say this.... 10, 50, 1000 HDDs? When people say xx HDDs failed are they including experiences from last centrurys HDD technology? How can a data set this small disprove the findings of a much more laborious and vastly larger experiment.? As i said in this very beginning until I read this paper and the follow ons I would have agreed 100% with what people are saying here but from a scientific veiwpoint the evidence is very strong.
May 4, 200818 yr One of the points missed here is that they are saying "cold" hard drives. Not warm hard drives operating within the optimum temperatures of 35c to 47c. Another point is the graphs show higher failure rates within infant mortality periods. Once you pass 6 months they seem to be good, thereafter around the 3yr mark they fail because of heat.
May 5, 200818 yr Playing devils advocate so i can understand better why this paper is incorrect... How do you KNOW heat caused a HDD to fail, surely in most cases it is actually "my HDD failed and it was hot therefore it broke because it was hot"? Isnt that like saying I crashed my car when it was sunny therefore the sun made me crash? In my scenario, the rest of the case (plastic/composite) melted around it... >.> ehe. Apparently computers don't like being locked inside small cabinets for multiple days.
Archived
This topic is now archived and is closed to further replies.