Google Releases Paper on Disk Reliability

oski4410 writes "The Google engineers just published a paper on Failure Trends in a Large Disk Drive Population. Based on a study of 100,000 disk drives over 5 years they find some interesting stuff. To quote from the abstract: 'Our analysis identifies several parameters from the drive's self monitoring facility (SMART) that correlate highly with failures. Despite this high correlation, we conclude that models based on SMART parameters alone are unlikely to be useful for predicting individual drive failures. Surprisingly, we found that temperature and activity levels were much less correlated with drive failures than previously reported.'"

Great

[true_hacker (969330)]

Excellent, i have been looking forward to thi *%)%*# DISK FAILURE

Re:

[Compholio (770966)]

Excellent, i have been looking forward to thi *%)%*# DISK FAILURE

That's what you get for logging into slashdot from Antarctica...

Re:Proprietary reporting

[spisska (796395)]

ps.. all their farm is ata/ide?

You really didn't read the article, did you? On page 3 (Section 2.2 Deployment Details), the authors state: "More than one hundred thousand disk drives were used for all the results presented here. The disks are a combination of serial and parallel ATA consumer-grade hard disk drives, ranging in speed from 5400 to 7200 rpm, and in size from 80 to 400 GB. All units were put into production in or after 2001. [...] The data used for this study were collected between December 2005 and August 2006."

What are you waiting for Google to tell you? Are you really accusing them of being evil because they did a study, described their methodology, detailed their results, presented their analyses, and published it all for anyone who is interested?

You describe their conclusions as:

Uselsess

But there is no contradiction at all if you are smart enough to understand. They are telling you that if SMART identifies a problem with a drive then it is very likely that drive will fail within 60 days. But in a sample of 100,000 drives, many drives will also fail that have not returned errors on SMART scans. Thus SMART is a reliable indicator of impending failure but is not a silver bullet that can recognize and predict all failures before they happen.

Next time you have access to 100,000 hard drives, can analyze patterns of failure among them, can use those failures as a benchmark against which to measure analysis tools, and can come up with better recommendations for predicting failure than this study, then by all means let us know. But if you're looking for Microsoft or Western Digital or Seagate or Yahoo to perform and publish this kind of study for free, I think you may be waiting a good long while.

So SMART is specific, but not sensitive.

[spineboy (22918)]

To me it's useful - if I get a SMART warning, then I'm definitely backing up my drive and will replace it before it croaks.

Sensitivity/specificity always presents a balancing act of testing, and they are usually in a push/pull relationship. If you make a test too sensitive, then you get too many false positives, and wind up over treating something (i.e. the test says it might fail so you replace the drive even though it's not going to - a false alert)

If you make the test too specific, then usually you wind up decreasing it's sensitivity, or ability to detect something. Now you get false negatives, so when the test works, you can be sure that it's accurate, but it always doesn't detect the problem.

What you want to know is the Positive Predictive Value PPV, which is determnined by the formula PPV=TP/(TP+FP). TP= true positives, FP = false positives
Also useful is the Negative Predictive Value NPV, or this formula NPV=TN/(FN+TN) where TN = true negative, FN = false negative.

What information these give are as such. If a test is positive (i.e. the drive temperature is >80 C), then it accurately will predict that the drive will fail. If the test is negative (drive temp 40 C0 then it accurately predicts that the drive is ok.

Re:Proprietary reporting

[T-Ranger (10520)]

They are hardly trade secrets. Google isn't in the hardware business. There are only so many patterns of disk usage on can have, and knowing what pattern Google has would hardly be useful to figure out how they did anything that they do. At least, to any level of detail useful enough to copy.

The amount of positive press they get from these types of releases easily justifies the effort to polish internal reports up to a publication standard. By releasing these types of papers, others may change their buying habits, which in turn will change the products sold. Google may believe that these types of papers would cause shame, not from individual manufacturers, but the industry in a whole, and thus cause better products to be produced.

Re:

[Toba82 (871257)]

It is well known that google uses commodity hardware. SCSI is not commodity, although I'm sure at least some of their servers are high end.

Did they ever name the brands?

[SuperKendall (25149)]

They stated at one point in the document that some brands did have higher failure rates than others - yet I somehow missed any mention or ranking of brands. Did anyone else find that data?

That would be corporate dynamite

[Traf-O-Data-Hater (858971)]

I noticed this too. If a Google-sanctioned report had charts of which brands were more reliable, this would do serious damage to the brands that didn't perform so well. No wonder they sidestepped the whole issue!

Re:That would be corporate dynamite

[MrZaius (321037)]

It's no wonder that Google sidestepped the issue, but, if you assume they purchase primarily from the manufacturers that are more reliable, perhaps those manufacturers will begin to gloat and publish numbers about their Google contracts, if this study gains traction.

Re:That would be corporate dynamite

[Antique Geekmeister (740220)]

I'm confident that Google is fairly drive agnostic: you just can't run distributed networks that large and stay locked into a single vendor. And given that even reliable vendors have disasters like the IBM Deskstar drives some years ago, and given the remarkable growth of drive sizes over time, there's just not much point for them in buying the extremely stable but vastly more expensive hardware. They've foubtless learned that hardware flexibility provides valuable software flexibility.

They do say that "vintage" matters

[Joce640k (829181)]

The report does say that "vintage" matters, ie. that "Past performance is not a reliable indicator of future development".

Manufacturers have good years and bad years. The writers don't want to damn a company because it had a couple of bad years during this time period.

Still, it's a bummer that the single most important factor goes unpublished. Even if it could cause a panic I'm sure there's some useful information in there (eg. a company to avoid like the plague).

Re:That would be corporate dynamite

[EonBlueTooL (974478)]

Google:Organizing all the world's information and making it universally accessible and useful(unless it could be troublesome)

Re:

[devilspgd (652955)]

Organizing and making accessible information which is already available is one thing, producing information is completely different.

Re:That would be corporate dynamite

[Jah-Wren Ryel (80510)]

Google:Organizing all the world's information and making it universally accessible and useful(unless it could be troublesome)

Old Google Motto: Don't do anything evil.
New Google Motto: Don't get into trouble.

You can get IDE/SATA drives FAILURE RATES Here

[Augur (62912)]

One of largest retailers in Russia (and maybe in Europe - more than 300 terminals for orders in person at ex-factory building, busy 24/7) "Pro Sunrise" released information on failure rates of major components (CPU, Videocards, motherboards, IDE/SATA, etc) of PC they sold for Q1-Q2 of 2005.

http://pro.sunrise.ru/articletext.asp?reg=30&id=28 3 [sunrise.ru] - the article (in russian, but diagrams are self-explanatory).

http://pro.sunrise.ru/docs/30/image001.gif [sunrise.ru] - IDE/SATA (3.5" formfactor)

http://pro.sunrise.ru/docs/30/image002.gif [sunrise.ru] - HDD (2.5" notebook formfactor)

In short, most returns are for Maxtor brand. Lowest - IBM/Hitachi.

Toshiba is worst in 2.5", and Seagate is best.

The chance to be blown are between 1/20 (Maxtor) to 1/70 (Hitachi).

Re:That would be corporate dynamite

[gbjbaanb (229885)]

When a friend broke down, she asked the breakdown man who came what were the most reliable cars. He said he wasn't allowed to comment but that "he carried no honda parts". I guess the same thing applies here - Google won't say, they'd get sued.

On the other hand, hard drives change so much that this year's model will be totally different design and mechanics than next years, so blaming (say) IBM for its crappy deskstar range should not be reason to blame their (ok, Hitachi's) current line.

If you do want to know more about which drives are best - check out storeagereview [storagereview.com] and enter details of your drives to their reliability database.

Re:Did they ever name the brands?

[iminplaya (723125)]

FTA:However, in this paper, we do not show a
breakdown of drives per manufacturer, model, or vintage
due to the proprietary nature of these data.

But, of course.

Re:

[ryturner (87582)]

It would be useful to you and me. But it is not useful to google to release that information.

Re:Did they ever name the brands?

[HUADPE (903765)]

There are several good reasons to not release the brand names. First, while the sample size is huge, the sample size for a particular model of a particular brand might not be. If they only happened to have 10 of one particular model, and one failed within a month, then 10% fail within a month, but it could just be a fluke. Second, liability. This wasn't a controlled test, it was done live within the Google servers (presumably). Whoever is on the bottom of the list could very well sue Google for libel. Without merit? Probably, but they might eke a few million in a settlement out of them. Google can't appear to be doing evil after all.

Re:Did they ever name the brands?

[iminplaya (723125)]

Well, obviously you're not a lawyer :-) Otherwise you would know the answer.

Re:

[Xross_Ied (224893)]

They didn't include any data at all about brands.

They should have done brand analysis (without naming the brand) and also rpm analysis.

From the article..

3.2 Manufacturers, Models, and Vintages
Failure rates are known to be highly correlated with drive
models, manufacturers and vintages [18]. Our results do
not contradict this fact. For example, Figure 2 changes
significantly when we normalize failure rates per each
drive model. Most age-related results are impacted by
drive vintages. However, in this paper, we do

Re:

[drmerope (771119)]

No. They explicitly said they would not disclose that... which is a shame because that is probably the only interesting bit of information. The question that really needs to be studied is what distinguishes good drives from bad. This would probably involve disassembling drives of various 'vintages, models, manufacturers' and trying to pin down the relevant details. That way when new hard-drives get released, reviewers can pull them apart and judge them on something other than read/write performance, hea

Re:Did they ever name the brands?

[Prof.Phreak (584152)]

At the very least, they could've named brands X, Y, Z, etc., and provided the numbers for those. Would be interesting if the differences are more than marginal.

Re:

[Schraegstrichpunkt (931443)]

They explicitly said they would not disclose that... which is a shame because that is probably the only interesting bit of information.

What? So the part about which variables are correlated with drive failures (which is what the report was about) wasn't interesting to you? Too bad.

Re:

[repvik (96666)]

"However, in this paper, we do not show a breakdown of drives per manufacturer, model, or vintage due to the proprietary nature of these data." (From TFA)

Translation

[jd (1658)]

"We don't want to be sued to within an inch of our lives by certain very wealthy brands, due to US law allowing manufacturers to prohibit unfavourable reviews."

Ideally, they would have formatted the text to spell out the names of the brands if you take the first letter of every Nth word, or some specific column of text. (Or maybe they have...)

Re:Translation

[David Price (1200)]

More likely: "We buy millions of dollars worth of drives each year, and our buying decisions are driven in part by the reliability data that we collect. If we told everyone what kind of drives work best, more people would buy those drives, driving up the price that we pay."

Re:Translation

[the_womble (580291)]

Another translation: Our competitors buy millions of dollars worth of drives as well. We are not going to help them avoid the duff ones.

Re:Translation

[spisska (796395)]

Another translation:

We're not so bloody stupid to believe that our competitors are standing in the aisle of Circuit City and scratching their head over whether to buy a Seagate or WD drive.

We know that our competitors all have their own metrics and their own relationships with manufacturers and frankly, we don't care. We know our competitors also measure these things, and we're not telling them anything they don't already know.

We aren't particularly worried about saying that some drives fail, because everyone who cares already knows that some drives fail. Everyone whose job it is to know which drives fail first already knows that as well.

But we're not going to tell you which brand fails at a higher rate than normal because we don't need a lawsuit that would cost us a lot of money but in the end would only confirm what the people who need to know these things already know.

We will, on the other hand, describe the tests we ran, our methodology, our results, and our analyses. We do this just for kicks and we hope you can learn something from the results.

And we hope you have a nice day.

Re:

[bendodge (998616)]

How did that get modded insightful? When there is more demand the price goes down, not up!

Re:Did they ever name the brands?

[by Anonymous Coward]

They would have released that data, but it was saved on a Maxtor.

Google had this paper ready a year ago

[by Anonymous Coward]

But the disk it was on failed.

Conclusion

[llZENll (545605)]

This is awesome, but the conclusion of such an interesting study leaves a lot to be desired. FTA...

"In this study we report on the failure characteristics of consumer-grade disk drives. To our knowledge, the study is unprecedented in that it uses a much larger population size than has been previously reported and presents a comprehensive analysis of the correlation between failures and several parameters that are believed to affect disk lifetime. Such analysis is made possible by a new highly parallel health data collection and analysis infrastructure, and by the sheer size of our computing deployment.

One of our key findings has been the lack of a consistent pattern of higher failure rates for higher temperature drives or for those drives at higher utilization levels. Such correlations have been repeatedly highlighted by previous studies, but we are unable to confirm them by observing our population. Although our data do not allow us to conclude that there is no such correlation, it provides strong evidence to suggest that other effects may be more prominent in affecting disk drive reliability in the context of a professionally managed data center deployment.

Our results confirm the findings of previous smaller population studies that suggest that some of the SMART parameters are well-correlated with higher failure probabilities. We find, for example, that after their first scan error, drives are 39 times more likely to fail within 60 days than drives with no such errors. First errors in reallocations, offline reallocations, and probational counts are also strongly correlated to higher failure probabilities. Despite those strong correlations, 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. This result suggests that SMART models are more useful in predicting trends for large aggregate populations than for individual components. It also suggests that powerful predictive models need to make use of signals beyond those provided by SMART."

Similar paper

[reset_button (903303)]

I was at the talk, and it was very interesting. CMU also had a paper (PDF) [cmu.edu] about disk failures in the same conference (in fact, they presented one after the other).

and in the meanwhile...

[pedantic bore (740196)]

... at the same conference, Bianca Schroeder presented a paper [cmu.edu] disk reliability that developed sophisticated statistical models for disk failures, building on earlier work by Qin Xin [ucsc.edu] and dozen papers by John Elerath... [google.com]

C'mon, slashdot. There were about twenty other papers presented at FAST this year. Let's not focus only on the one with Google authors...

Re:and in the meanwhile...

[oGMo (379)]

While at a glance, it may seem like this is simply "the latest thing google did," and... let's be honest, given the editor in question... this was most likely the reason it made the front page. But while Bianca Shroeder's report, for instance, uses statistics from various unnamed sources and for various unnamed uses, the Google report is interesting because we know exactly where it's coming from and what it's being used for.

Of course, a truly insightful story would have taken this opportunity to compare Google's findings with the others and report on that.

Temperature conclusion

[phasm42 (588479)]

Their statistics on temperature seem very unusual. I'm surprised they didn't explore this more. For example, is the high failure rate associated with low temperatures because the drives were more likely to be inactive due to failure?

Re:

[Chalex (71702)]

The chart implies that the "optimal" operating drive temperature is 35-45 Celsius. Drive temperatures below room temperature (below 22 Celsius) is probably not a scenario that drive manufacturers optimise for.

Re:Temperature conclusion

[gnu-sucks (561404)]

My guess is this graph on temperature distribution is more or less a graph of temperature sensor accuracy. I can't imagine that drives at 50C had the lowest failure rate.

While this would require a more laboratory-like environment, a dozen drives of each type and manufacture could have been sampled at known temperatures, and a data curve could have been established to calibrate the temperature sensors.

There are lots of studies out there where drives were intentionally heated, and higher degrees of failure were indeed reported (this is mentioned in the google report too). So the correlation is probably still valid, just not well-proven.

Re:

[bouis (198138)]

If hard drives are anything like car engines [especially those made with iron and aluminum], the designers have taken the standard operating temperature into account in the design. The parts of varying composition fit together best at the right temperature, and temperatures higher or lower result in damage or accelerated wear.

This is why, if you want your engine to last, you should let your car warm up before driving it hard.

Lower temp == higher failure rates

[flyingfsck (986395)]

To my mind the most significant piece of info: "The gure shows that fail- ures do not increase when the average temperature in- creases. In fact, there is a clear trend showing that lower temperatures are associated with higher failure rates. Only at very high temperatures is there a slight reversal of this trend."

Proprietary makes sense here

[Mammothrept (588717)]

"...we do not show a breakdown of drives per manufacturer, model, or vintage due to the proprietary nature of these data."

Litigation avoidance may be a consideration here but why not take Google at their word? Google is a search company that buys lots of hard drives. Based on their own internal research, they have developed information about which hard disk models and/or manufacturers are shite.

Yahoo is also a search company that buys lots of hard drives. Why should Google give that hard drive reliability information to you, me and Yahoo for free? Let Yahoo/Excite/MSN and the competitors figure it out for themselves.

Yeah, sure I'd like to have access to Google's data the next time I'm in the market for a hard drive but I won't hold a grudge against them if they don't do my consumer research for me. On the other hand, whereinafuck is the data from Tom's Hardware Guide, Anandtech, Consumer Reports and all the other reviewer and consumer sites? If someone doesn't have a handy link to their results, I'll see if I can google something up:

http://www.google.com/search?hl=en&safe=off&client =firefox-a&rls=com.ubuntu%3Aen-US%3Aofficial&hs=tq y&q=hard+drive+reliability+research+brands++manufa cturers+models&btnG=Search [google.com]

This speaks volumes.

[greenguy (162630)]

Google releases a paper on disk reliability.

How many drives really

[hankwang (413283)]

The paper claims "more than 100 thousand drives". But the nice thing is that you can derive the actual number from the error bars, for example those in figure 4. The data should be governed by Poisson statistics, which means that the standard deviation in the counts is equal to the square root of the count. However, their error bars seem to be about a factor 2 larger than the standard deviation, because normally around 68% of the data points should lie within one standard deviation from the "smooth curve". Let's assume the error bars are 95% confidence intervals, i.e. 2 standard deviations.

Look at the data for 20 to 21 C. It tells you that it represents a fraction 0.0135 of their total drive population, with an average failure rate of 7 +- 0.5 %. Following the reasoning above, this 7% should represent 784+-28 drives. Since these represent 7% of 1.35% of the total number of drives, we can derive that the total number of drives is 784/0.07/0.0135 = 830,000 drives. Trying the same thing for 30 to 31 C gives 826,000 drives, which seems fairly consistent.

So can we assume that Google has deployed 830,000 hard disk drives since 2001? How many servers do they have now?

Re:Hmm

[by Anonymous Coward]

Didn't read the article? (Check)
Didn't read the summary? (Check)

Congratulations, you're not officially a slashdot regular!

Re:Hmm

[pugugly (152978)]

Didn't check the typing? (Check)

Congratulations, you're now officially a slashdot regular! - Pug

Re:Hmm

[by Anonymous Coward]

There are several SMART signals which are highly correlated with drive errors, but the authors note that 56% of the failed drives had no occurrences of these highly correlated errors. Even considering all SMART signals, 36% of failed drives still had no SMART signals reported.

So, if you have errors in those highly correlated categories your drives are probably going to fail, but if you do not have errors in these categories your drives can still fail.

Re:Samsung!

[mollymoo (202721)]

In summary: Your statistical analysis on a sample size of one showed a 100% failure rate, so Samsung are crap. You found some other people also had failed Samsung drives, so Samsung are crap.

Search the net and you will find people ranting about Seagate drives failures, Western Digital drive failures, IBM drive failures, Maxtor drives failures and failures of drives made by companies neither of us have even heard of. You won't find many, if any, reports of recent failures with 8" floppy drives though, so I suggest you use one of those. They must be more reliable, right?

Re:Hmm

[jemenake (595948)]

So if the article summary is correct does it even matter if the consumer desktop pc has SMART enabled or not?

Well, I was a little disappointed by the article. They looked at a lot of different SMART categories and they looked at the different ages of the drives, but they didn't delve into the different types of failures. I get about 1 "I think my drive crashed and I was hoping you could recover it" call per month and I see a variety of failure types. Probably the most common ones I see now are ones where something has gone wrong with the control circuits/mechanism and not the media itself. For example, something can go wrong with the motor that spins the platters, or you can seize the bearings for the head traversal, etc. I've even seen some where a chip on the controller board literally popped when it got too hot. These aren't going to be detected by SMART... I don't know what would predict failures like that.

The article states that, in about half of the failures, there were no SMART warnings at all. Okay, but what was the breakdown in the kinds of failures of these unpredicted ones? If they were all spindle motor and head traversal failures, then you can't blame SMART for that. If it turns out that SMART gave warnings for 95% of all failures that were media-degradation related (like bad sectors, etc... where the drive still talks to your machine properly, and just can't get the data you want), then I'd say SMART is pretty darn useful.

But, alas, I didn't see any breakdown for failure type....

Re:OS X SMART tool?

[kimvette (919543)]

http://sourceforge.net/projects/smartmontools [sourceforge.net]

Not exactly point & click but it'll do.