25

First of all, I think everyone knows that hard drives fail a lot more than the manufacturers would like to admit. Google did a study that indicates that certain raw data attributes that the S.M.A.R.T status of hard drives reports can have a strong correlation with the future failure of the drive.

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.

Seagate seems like it is trying to obscure this information about their drives by claiming that only their software can accurately determine the accurate status of their drive and by the way their software will not tell you the raw data values for the S.M.A.R.T attributes. Western digital has made no such claim to my knowledge but their status reporting tool does not appear to report raw data values either.

I've been using HDtune and smartctl from smartmontools in order to gather the raw data values for each attribute. I've found that indeed... I am comparing apples to oranges when it comes to certain attributes. I've found for example that most Seagate drives will report that they have many millions of read errors while western digital 99% of the time shows 0 for read errors. I've also found that Seagate will report many millions of seek errors while Western Digital always seems to report 0.

Q: How do I normalize this data? Is Seagate producing millions of errors while Western digital is producing none? Wikipedia's article on S.M.A.R.T status says that manufacturers have different ways of reporting this data.

Here is my hypothesis:

I think I found a way to normalize (is that the right term?) the data.

Seagate drives have an additional attribute that Western Digital drives do not have (Hardware ECC Recovered). When you subtract the Read error count from the ECC Recovered count, you'll probably end up with 0. This seems to be equivalent to Western Digitals reported "Read Error" count. This means that Western Digital only reports read errors that it cannot correct while Seagate counts up all read errors and tells you how many of those it was able to fix.

I had a Seagate drive where the Read error count was less than the ECC Recovered count and I noticed that many of my files were becoming corrupt. This is how I came up with my hypothesis. The millions of seek errors that Seagate produces are still a mystery to me.

Please confirm or correct my hypothesis if you have additional information.

Here is the smart status of my western digital drive just so you can see what I'm talking about:

james@ubuntu:~$ sudo smartctl -a /dev/sda
smartctl version 5.38 [x86_64-unknown-linux-gnu] Copyright (C) 2002-8 Bruce Allen
Home page is http://smartmontools.sourceforge.net/

=== START OF INFORMATION SECTION ===
Device Model:     WDC WD1001FALS-00E3A0
Serial Number:    WD-WCATR0258512
Firmware Version: 05.01D05
User Capacity:    1,000,204,886,016 bytes
Device is:        Not in smartctl database [for details use: -P showall]
ATA Version is:   8
ATA Standard is:  Exact ATA specification draft version not indicated
Local Time is:    Thu Jun 10 19:52:28 2010 PDT
SMART support is: Available - device has SMART capability.
SMART support is: Enabled

=== START OF READ SMART DATA SECTION ===
SMART overall-health self-assessment test result: PASSED

SMART Attributes Data Structure revision number: 16
Vendor Specific SMART Attributes with Thresholds:
ID# ATTRIBUTE_NAME          FLAG     VALUE WORST THRESH TYPE      UPDATED  WHEN_FAILED RAW_VALUE
  1 Raw_Read_Error_Rate     0x002f   200   200   051    Pre-fail  Always       -       0
  3 Spin_Up_Time            0x0027   179   175   021    Pre-fail  Always       -       4033
  4 Start_Stop_Count        0x0032   100   100   000    Old_age   Always       -       270
  5 Reallocated_Sector_Ct   0x0033   200   200   140    Pre-fail  Always       -       0
  7 Seek_Error_Rate         0x002e   200   200   000    Old_age   Always       -       0
  9 Power_On_Hours          0x0032   098   098   000    Old_age   Always       -       1468
 10 Spin_Retry_Count        0x0032   100   100   000    Old_age   Always       -       0
 11 Calibration_Retry_Count 0x0032   100   100   000    Old_age   Always       -       0
 12 Power_Cycle_Count       0x0032   100   100   000    Old_age   Always       -       262
192 Power-Off_Retract_Count 0x0032   200   200   000    Old_age   Always       -       46
193 Load_Cycle_Count        0x0032   200   200   000    Old_age   Always       -       223
194 Temperature_Celsius     0x0022   105   102   000    Old_age   Always       -       42
196 Reallocated_Event_Count 0x0032   200   200   000    Old_age   Always       -       0
197 Current_Pending_Sector  0x0032   200   200   000    Old_age   Always       -       0
198 Offline_Uncorrectable   0x0030   200   200   000    Old_age   Offline      -       0
199 UDMA_CRC_Error_Count    0x0032   200   200   000    Old_age   Always       -       0
200 Multi_Zone_Error_Rate   0x0008   200   200   000    Old_age   Offline      -       0

Edit: Here is the Seagate drive that I was talking about that was causing data corruption. This data is from HDTune.

HD Tune: ST3250623A Health

ID                               Current  Worst    ThresholdData       Status   
(01) Raw Read Error Rate         45       38       6        77882492   Ok       
(03) Spin Up Time                99       98       0        0          Ok       
(04) Start/Stop Count            100      100      20       640        Ok       
(05) Reallocated Sector Count    100      100      36       0          Ok       
(07) Seek Error Rate             85       60       30       359872048  Ok       
(09) Power On Hours Count        94       94       0        6028       Ok       
(0A) Spin Retry Count            100      100      97       0          Ok       
(0C) Power Cycle Count           100      100      20       689        Ok       
(C2) Temperature                 25       55       0        25         Ok       
(C3) Hardware ECC Recovered      50       47       0        201555081  Ok       
(C5) Current Pending Sector      100      100      0        0          Ok       
(C6) Offline Uncorrectable       100      100      0        0          Ok       
(C7) Ultra DMA CRC Error Count   200      199      0        1          Ok       
(C8) Write Error Rate            100      253      0        0          Ok       
(CA) TA Counter Increased        100      253      0        0          Ok       

Power On Time         : 6028
Health Status         : Ok

The fact that the Hardware ECC Recovered is larger than the Raw Read Error Rate is counter intuitive in my opinion.

This is what I've found to be a "normal" seagate drive where the ECC Recovered matches the Raw Read Error Rate:

HD Tune: ST380011A Health

ID                               Current  Worst    ThresholdData       Status   
(01) Raw Read Error Rate         62       46       6        79986164   Ok       
(03) Spin Up Time                98       98       0        0          Ok       
(04) Start/Stop Count            100      100      20       6          Ok       
(05) Reallocated Sector Count    100      100      36       0          Ok       
(07) Seek Error Rate             83       60       30       210309663  Ok       
(09) Power On Hours Count        93       93       0        6516       Ok       
(0A) Spin Retry Count            100      100      97       0          Ok       
(0C) Power Cycle Count           99       99       20       1325       Ok       
(C2) Temperature                 25       52       0        25         Ok       
(C3) Hardware ECC Recovered      62       46       0        79986164   Ok       
(C5) Current Pending Sector      100      100      0        0          Ok       
(C6) Offline Uncorrectable       100      100      0        0          Ok       
(C7) Ultra DMA CRC Error Count   200      188      0        18         Ok       
(C8) Write Error Rate            100      253      0        0          Ok       
(CA) TA Counter Increased        100      253      0        0          Ok       

Power On Time         : 6516
Health Status         : Ok

EDIT:

I want to clarify that I know that Google generally considers S.M.A.R.T useless. I know that everyone should backup their data. I am however in the business of fixing other peoples computers. Most people do not have backups or have RAID. It is not cost effective for corporations to troubleshoot hard drives, so they just run them on a RAID until they die. I find it useful in my line of work to check the SMART status of the hard drive. It takes like 30 seconds. If I am lucky enough for a bad drive to show a hint of failure such as scan errors or reallocated sectors, I know to get the drive the heck out of there. If no such hint exists, I'll probably spend many hours troubleshooting slowness and data corruption until I finally find that the hard drive is bad.

I'm just trying to fine tune this process.

3
  • There is smart based information in the administration menu under (I believe) disks management. It may have additional abilities over smartctl, but I haven't used it in a while and don't have it in front of me.
    – Jarvin
    Jun 14, 2010 at 20:54
  • 1
    @Dan Hi Dan, I'm not sure what windows tool you are talking about. Can you clarify?
    – James T
    Jun 16, 2010 at 19:18
  • The problem with SMART is that it is a bit of a misnomer; there is no actual intelligence in it, only a few equations (probably not even heuristics). All it can do is monitor itself and report the numbers, that’s all. For example, I have a drive that had a poorly connected power cord, which caused it to turn on and off very quickly several times (making a “click-of-death” sound). I reseated the connector, so it works smoothly now, but due to the temporary (fixable) failure that one time, it has now permanently recorded a RRER event in the SMART, making look like it’s failing.
    – Synetech
    Jan 1, 2012 at 0:03

6 Answers 6

15

It does appear that different manufacturers use SMART values for sometimes radically different things, as you can see here:

My hard disk(s) in ReadyNAS is reporting high SMART Raw Read Error Rate, Seek Error Rate, and Hardware ECC Recovered. What should I do?

Seagate uses these SMART fields for internal counts, so this is a known issue with Seagate disks. Look for abnormal counts in other fields, especially Reallocated Sector Ct and ATA Error Count.

So when it comes to your actual question ...

If I am lucky enough for a bad drive to show a hint of failure such as scan errors or reallocated sectors, I know to get the drive the heck out of there. If no such hint exists, I'll probably spend many hours troubleshooting slowness and data corruption until I finally find that the hard drive is bad.

I'd say a good rule of thumb is, you can only expect SMART settings to be comparable within the same drive manufacturer, and maybe even the same drive model!

So when you're looking at diagnosing those SMART counts, keep that in mind... one manufacturer's "read error retry count" may mean something totally different than another manufacturer's. Sad but true. :(

0
18
+50

Okay, first of all I disagree with your premise.

Google did a study that indicates that certain raw data attributes that the S.M.A.R.T status of hard drives reports can have a strong correlation with the future failure of the drive.

In fact they found the opposite:

...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.

Secondly, SMART thresholds are not standardised. The firmware on the drive itself will flag an attribute as being "pre-failure", but the raw values are meaningless to the user. For example, Seagate says:

Various attributes are being monitored and measured against certain threshold limits. If any one attribute exceeds a threshold then a general SMART Status test will change from Pass to Fail.

The SMART values that might be read out by third-party SMART software are not based on how the values may be used within the Seagate hard drives. Seagate does not provide support for software programs that claim to read individual SMART attributes and thresholds. There may be some historical correctness on older drives, but new drives, no doubt, will have incorporated newer solutions, attributes and thresholds.

tl;dr Summary:

Raw SMART values are almost meaningless, as different manufacturers use them in different ways and have different thresholds etc. The drive firmware itself will tell you when it is in "pre-failure"... or it might not, SMART really isn't very reliable.

Do regular backups!

15
  • 1
    Based on your comments it does not seem like you read my whole post. This is why I put in all the background information and quotes. You quoted Google but only a very select part of it. If you read the part just before your quote... it says that some attributes have a strong failure correlation.... such as reallocated sector counts. The manufacturers do not report their drives as being in a pre-failure state after one reallocated sector. This clearly indicates that you can get a better indication of the health of the drive by looking at the raw data.
    – James T
    Jun 16, 2010 at 18:33
  • I'd also like to add that my seagate drive was corrupting my data and the raw data values were noticeably different from what I've learned to be healthy drives. Clearly something is wrong with where the manufacturer sets the threshold.
    – James T
    Jun 16, 2010 at 18:35
  • I think you need to re-read my post and link. Raw SMART values are not reliable indicators of anything. The Google report does not say that "some attributes have a strong failure correlation". What it does say is that despite the fact that "after their first scan error, drives are 39 times more likely to fail within 60 days than drives with no such errors", less than 15% of the failed drive population had any Scan Errors. Is it a reliable indicator if it is right 15% of the time?
    – sml
    Jun 16, 2010 at 19:24
  • 1
    @scottl I'm not sure where you got your 15% from. I did not see that in the article. Even if only 15% of their drives had scan errors... they found that a drive with scan errors is 39 times more likely to fail in 60 days. This does not mean that your drive will not fail unless you have scan errors. This just means that if you do have a scan error... your hard drives remaining life is probably short. Have you ever taken statistics? I found it very useful.
    – James T
    Jun 16, 2010 at 19:41
  • 1
    smartmontools FAQ says: The raw SMART attributes (temperature, power-on lifetime, and so on) are stored in vendor-specific structures. Sometime these are strange. Hitachi disks (at least some of them) store power-on lifetime in minutes, rather than hours (see next question below). IBM disks (at least some of them) have three temperatures stored in the raw structure, not just one. And so on.
    – sml
    Jun 16, 2010 at 19:41
4

I'm not exactly sure what the question is that you're asking. You seem to have the whole question and answer rolled up into one but...

Have you compared the hard drive metrics to those given from SeaTools

It's Seagate's standard hardware diagnostic tool and AFAIK the most commonly used HDD diagnostic tool.

Don't be surprised if you find that the tools report unfavourable results about their competitors. The tools generally work with HDDs of all manufacturers but that doesn't mean that they have make their competitors look good while doing.

Haven't you ever heard the joke, "99.99% of all statistics are true except, of course, this statistic".

5
  • 1
    Yeah... it is a bit confusing. I basically put in all the background information that I am familiar with before the question and all my tests and conjectures after the question. Here is my question "How do I normalize this data?". Basically.. how do I make all the data attributes from one manufacturer mean the same thing as the data attributes from another manufacturer so I can accurately compare them.
    – James T
    Jun 15, 2010 at 22:54
  • @James You can try to collect data from as many difference as possible and figure out how each if interpreting the data differently from one another. They may all be reporting correct data, they may just be interpreting it in a different manner like you pointed out. That's why I added the statistics quote... Just because the data is good, doesn't mean the interpretations is. Jun 15, 2010 at 23:16
  • 2
    Yup, that is what I have done. I've checked over 70 different hard drives and the large difference in seek errors and read errors are the attributes that stuck out to me. I have a guess that for seagate drives, read errors have some kind of relationship with hardware ecc recovered. I'm not exactly sure what that relationship is. I was hoping someone here could tell me. I was also hoping someone could tell me why seagate drives have huge seek error counts while western digital always seems to have zero.
    – James T
    Jun 15, 2010 at 23:23
  • @James Maybe somebody will come along with a better answer... My honest guess is, Western Digital probably doesn't follow the exact S.M.A.R.T spec. That's the problem with hardware standards, they're great selling points but there's always a few manufacturers that will market all the benefite without following the full specification. Jun 15, 2010 at 23:29
  • Yup the deviation from the standard is what I figured and what the wikipedia article suggests. I'd like to know how they differ so that I might be able to properly compare the two manufacturers (and possibly others). Thanks for the comments Evan. Hopefully this clarifies the question for others too.
    – James T
    Jun 15, 2010 at 23:37
3

In the physical reality of hard drive internals, all brands of hard drives larger than 100MB will have a lot of physical read errors. Most of those are safely corrected by ECC, some (hopefully very few) are wrongly corrected by ECC and the rest (few but more than the wrong corrections) are reported back to the computer as failed read and should also make the drive automatically relocate the bad sector.

In addition to correcting raw read errors, ECC also corrects reads that the hardware thought were OK, but the returned bits were slightly wrong. Thus ECC corrected might be "raw read failed but fixed by ECC + raw read succeeded but was wrong and got fixed by ECC".

Thus two interpretations of the data seem possible:

A. non-Seagate drives do not include the ECC corrected read errors in the "raw read error count", only the unfixable errors.

B. Seagate considers it a read error if ECC finds something wrong with the data even if the low level circuit did not notice, others don't.

Normalization will be very different depending on which theory (A or B) is right.

2
  • > should also make the drive automatically relocate the bad sector. Then what is the relation between the Uncorrectable Sector Count Relocated Event Count and Current Pending Sector Count fields? Wouldn’t it increase current, then either relocated or uncorrectable? Why would it be uncorrectable? If it tried to remap a bad sector and it failed (ie, the spare sector is bad), then shouldn’t it try remapping to a different spare sector? it’s not a tire that it only has one spare.
    – Synetech
    Dec 31, 2011 at 23:54
  • 100 MB? Do you mean 100 GB? Dec 7, 2013 at 17:07
0

You can look at the Seagate specification at: http://t1.daumcdn.net/brunch/service/user/axm/file/zRYOdwPu3OMoKYmBOby1fEEQEbU.pdf.

The millions of seek errors that Seagate produces are still a mystery to me.

that's because probably they aren't errors. From spec:

**Attribute ID 7: Seek Error Rate**
Monitor seeks requiring one or more retries. Exclude calibration seeks and seeks in system area.
Normalized Seek Error Rate = 10 * log10(SeekCount / SeekErrors) which is only updated when
SeekCount is in the range 106 to 109. The counts are cleared when SeekCount = 109 . (Evaluates
to a value from 1 to 100).
**Raw Usage**
Raw [3 – 0] = Number of seeks
Raw [5 – 4] = Number of seek errors

ID                               Current  Worst    ThresholdData             
(07) Seek Error Rate             85       60       30       359872048  Ok       

359872048 > 0x000015733630 > 0 errors, 359872048 seeks.

Also see: http://www.users.on.net/~fzabkar/HDD/Seagate_SER_RRER_HEC.html

-1

Raw SMART values for the 01 Raw Read Error Rate and 195 Hardware ECC Recovered are indeed almost meaningless to us. On Seagate drives they appear to contain a bitwise data structure, which I assume probably is derived from the cylinder/head/sector or an abbreviation of it. Using the Hard Disk Sentinel chart view it shows historical values which jump up and down. 6 peaks and valleys on one, over 20 on another, over the life of the drive, with raw data values values between 20 and 200 million. I expect it shows the most recent occurrence only.

This fits with your observation that when the 01 and 195 are in disagreement the drive has experienced trouble: a sector it had an error on, shown in the 01 Raw Read but not 195 Recovered. However, I'd also then expect a value in either Current Pending Sector, Reallocated Sector Count or Offline_Uncorrectable, which isn't shown in your example.

Other raw values such as 09 Power On Time are in hours on Seagate, and quite transparent.

1
  • To the person who gives a "this answer is not useful" click, please add a comment so that I can understand where I went wrong, thank you. Jan 28, 2023 at 22:45

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .