Google's study on Hard Drives

[crazypyro]

http://hardware.slashdot.org/article.pl?sid=07/02/18/0420247&from=rss
This is where i read about it first,

http://216.239.37.132/papers/disk_failures.pdf
Googles findings. its a PDF



I found the read to be very interesting and resourceful when considering hard drive purchases. As the summarized article from slashdot says, i'm shocked that SMART isn't very smart, and heat issues don't necessarily mean drive failure.

Spreading some news, feel free to add your hate or love.

 

[mpjesse]

Dude... great find. Clearly temperatures play a very small role in disk failures. It looks like the best temperature to keep your drive at is between 38C and 43C, which is the average temp of a desktop case.

At very cold temperatures the culprit likely the difference in drive material temperature and ambient air temp. It probably puts a lot of structural stress on the materials. (I.E., exposing a very hot surface to very cold air increases contraction)

Another interesting statistic is the AFR (annual failure rate). At exactly 3 years a drive is most likely to fail. One wonders if this is a coincidence or part of a more intelligent design. Most drives have 3 year warranties.

Thanks for posting this. Definitely worth the read.

 

[mpjesse]

One thing I would like to point out though is that SMART doesn't do a horrid job. They concluded that if SMART reports any error in a drive, it has a high probability of failing.

So... with respect to an "early warning system" SMART works pretty well.

 

[mpjesse]

Why is no one discussing these findings?!

Come on guys...

 

[sandmanwn]

well its a nice study :?

the thing is the google environment is a great study for other companies to draw conclusions from. I could use this for failure predictions to anticipate yearly budgeting concerns for a business.

Your experienced user on the other hand is likely to reformat the drive twice a year/switch os's/drop the drive on the floor/experience power spikes/make good ole ID-10-T mistakes/use a raid array/etc... so the study really doesnt do much for your average Joe.

Your average user barely uses their computer and therefore the burden to the drive is minimal and is likely to exceed the 5 yr protection plan. Or they end up selling it off long before any of these issues come to play.

what more do you want from me

:lol:

 

[sandmanwn]

the most interesting part of the article was when they mentioned that the hard drive models and vintages play a major role. I believe it was section 3.2

Unfortunately they did not record that information. Which would have really been the most pertinent information in the whole article.

I much rather would have liked to of seen a breakdown of the models and failure rates than percentages.

 

[SomeJoe7777]

Al lot of the HDs manufactured in the 2001-2004 were of a trend where manufacturers were trying to make the drives cheaper and shaving the warranty. A lot of drives made back then were really crappy ... Maxtor DiamondMax 8's & 9's in particular.

We don't know exactly what Google bought, but I'm of the opinion that hard drive quality has come up since then. I believe the newer SATA Seagates & Western Digitals look & feel to be better quality that 3-5 year old models.

I thought the SMART analysis in the article was the most important thing. The fact that Google boiled it down to the 4 SMART parameters that matter (Scan Errors, Reallocated Sectors, Offline Reallocated Sectors, and Probational Errors) is important. It's good to know that ECC Corrected Reads and Seek Errors (particularly for Seagate drives) don't mean much.

 

[mpjesse]

the most interesting part of the article was when they mentioned that the hard drive models and vintages play a major role. I believe it was section 3.2

Unfortunately they did not record that information. Which would have really been the most pertinent information in the whole article.

I much rather would have liked to of seen a breakdown of the models and failure rates than percentages.

Google probably didn't want to risk busting out their hard drive partners. But yes, it would be very interesting to see which manufacturer was the most reliable.

 

[crazypyro]

the most interesting part of the article was when they mentioned that the hard drive models and vintages play a major role. I believe it was section 3.2

Unfortunately they did not record that information. Which would have really been the most pertinent information in the whole article.

I much rather would have liked to of seen a breakdown of the models and failure rates than percentages.

Google probably didn't want to risk busting out their hard drive partners. But yes, it would be very interesting to see which manufacturer was the most reliable.

I would've been happy to just see manufactures to failure rates. Don't need specifics, you can draw conclusions on your own based on who had the most that failed.

But depending on how they got the drives (retail bought, or manufacturer giving) i can see why they don't.

But i know WD 120 IDE drive is 4 1/2 yrs old still chumming alone nicely, but i should start looking for a replacement.

 

[zenmaster]

I actually started a thread a couple of days ago on the same topic.

The high drive failure rates for years 2 & 3 where just shocking to me.
I never would have guessed they were that high.

And yes, I would love to know the various manufacturer stats.

But I understand why they did not.
No need to piss of another company.

But it was very nice that the truth was told about the almost fraudulent MTBF rates posted by the manufacturers.

 

[Shadow703793]

Kind of old news btw. Saw this on BBC on Monday( I use FireFox and BBC live feeds). Here is the link.

 

[crazypyro]

I actually started a thread a couple of days ago on the same topic.

The high drive failure rates for years 2 & 3 where just shocking to me.
I never would have guessed they were that high.

And yes, I would love to know the various manufacturer stats.

But I understand why they did not.
No need to piss of another company.

But it was very nice that the truth was told about the almost fraudulent MTBF rates posted by the manufacturers.

[Post] (Msg. 1) Posted: Sun Feb 18, 2007 11:51 pm
Post subject: Google's study on Hard Drives

yea i posted it originally on the 18th, just one of them topics only the interested care about

 

[Lionhardt]

nice article

 

[mpjesse]

Kind of old news btw. Saw this on BBC on Monday( I use FireFox and BBC live feeds). Here is the link.

highspeed:

you may want to check the date this thread was originally posted.

 

[Ninjaz7]

Holy crap....it seems as though lately my hd's are droppin like flys 8O ...whats the best advice for a long lasting drive,performance is almost secondary due to the fact that striped array is sometimes more headache than what its worth ...just gimmie a hd that will at least hold up 5 yrs,or is there 1?

 

[zenmaster]

Try looking for drives that are "Enterprise" rated.

Example Seagate's 320gb drives come in the ES and AS models.
The ES ones are $5 more.

I buy those for extra safety.
Likely not much better, but does not hurt.

 

[Ninjaz7]

ya,sounds like a winner....I'm about to through in the towel with WD,they've ganked me out of enough cash and lost info, any other sugestions?

 

[zenmaster]

Well, I've heard that raptors may be the most reliable of all desktop hdds.

Look on the bright side, you don't have a laptop.
I can't tell you how many laptop HDDs I have lost over the years.

Bangs, Drops, etc... I guess are not so good for them :>>

 

[sandmanwn]

Holy crap....it seems as though lately my hd's are droppin like flys 8O ...whats the best advice for a long lasting drive,performance is almost secondary due to the fact that striped array is sometimes more headache than what its worth ...just gimmie a hd that will at least hold up 5 yrs,or is there 1?

Sounds like you need to figure out what you are doing to cause all these failures instead of blaming it on the manufacturer.

Not to flame you or anything but if you are loosing that many drives then the most likely culprit is the user not the drives. Sorry to put it that way but if their is a pattern of destruction then you have to start looking at the person and not the machine.

Either you arent getting enough airflow to the drives or perhaps not enough power from the PSU. Hard drive manufacturers dont design these drives to fail so if you are having repeated issues then there is most likely another culprit causing those problems.

 

[Ninjaz7]

If you can average the amount of time and the years an inteligent person would come to the conclusion that time has taken there toll on alot of computer hardware,particularly hard drives(I've had cpu's ,gpu's,mobo's last forever).And on a study that they have recently concluded,heat is not a huge factor for falures,I believe I picked up that info on this forum(unless you have a cpu bar-b-que).Out of the years that I have been useing WD products,they have not all served me well,be it used and quite a bit of cash dropped on new.I'm going to try a seagate http://www.newegg.com/Product/Product.asp?Item=N82E16822148140
and test the waters,maybe I've just had bad luck,maybe drives have advanced since my last purchase.Just keepin it real and reporting my finds. p. s.thanx for the info,striped raptors still sounds pretty sweet...What breed of HD has served you well,I wanna know.

 

[alandg46]

Wonder what the hd failure rate is on Macs and which hd's they use?

 

[Dante_Jose_Cuervo]

I think macs use Seagate drives now. Although I saw a WD in a MacPro last year, so who knows.

 

[sailer]

The high drive failure rates for years 2 & 3 where just shocking to me.
I never would have guessed they were that high.

But it was very nice that the truth was told about the almost fraudulent MTBF rates posted by the manufacturers.

This study seems strange to me. It makes is seem that either I've been very lucky, or the study itself is wrong. In the last 17 years that I've been buying hard drives, I've only had two of them fail, one a Maxtor and one because of weather damage which was totally unrelated to it being a bad drive. I've currently got a few drives running in my office computers that are between 5 to 7 years old. Oh yes, I do have all needed data backed up un various floppies and CDs.

One thing I did notice, was the five year failure rate was about 8%. The study didn't point out that meant that about 92% of the drives of that age didn't fail. If you added the total numbers of the failure rates, they would be about 38%, which is very high, but that still leaves 62% (or nearly two thirds) which kept on going. Ok, maybe I'm in that 62% which has no significate drive failures.

So I conclude that either this study is a bit on the alarmist side, or I'm misunderstanding its conclusions.

 

[crazypyro]

This study seems strange to me. It makes is seem that either I've been very lucky, or the study itself is wrong. In the last 17 years that I've been buying hard drives, I've only had two of them fail, one a Maxtor and one because of weather damage which was totally unrelated to it being a bad drive. I've currently got a few drives running in my office computers that are between 5 to 7 years old. Oh yes, I do have all needed data backed up un various floppies and CDs.

One thing I did notice, was the five year failure rate was about 8%. The study didn't point out that meant that about 92% of the drives of that age didn't fail. If you added the total numbers of the failure rates, they would be about 38%, which is very high, but that still leaves 62% (or nearly two thirds) which kept on going. Ok, maybe I'm in that 62% which has no significate drive failures.

So I conclude that either this study is a bit on the alarmist side, or I'm misunderstanding its conclusions.

This is why most of us would have liked to seen the models and manufacturer's used in this test, and what exactly they did to do the test. I'm like you i have a 5yr old western digital running flawlessly. Its been a in 3 computers, used as a main and used as storage, both of which were in systems that were never shut down (unless i had to or power went out).

 

[mpjesse]

One thing I did notice, was the five year failure rate was about 8%. The study didn't point out that meant that about 92% of the drives of that age didn't fail. If you added the total numbers of the failure rates, they would be about 38%, which is very high, but that still leaves 62% (or nearly two thirds) which kept on going. Ok, maybe I'm in that 62% which has no significate drive failures.

The flaw in your logic is that google likely upgrades its storage systems every 3-5 years. Meaning they don't have a hard drive in their data center older than 5 years because they've all been destroyed to pave way for newer, bigger HDDs. Hence, there's no study data for HDD's older than 5 years because they don't exist! ;-)