BackBlaze - HDD reliability stats for Q2 2015

[zMeul]

source: https://www.backblaze.com/blog/hard-drive-reliability-stats-for-q2-2015/

 

What Is A Failed Hard Drive?

For Backblaze there are three reasons a drive is considered to have “failed”:

The drive will not spin up or connect to the OS.

The drive will not sync, or stay synced, in a RAID Array (see note below).

The Smart Stats we use show values above our thresholds.

Note: Backblaze Vaults do not use RAID. Instead, we use our open-sourced implementation of Reed-Solomon encoding to replace the function of RAID. As a result, these drives are not subject to RAID-sync errors. RAID-sync failures are only applicable to stand-alone Storage Pods.

Takeaways

The 4TB drives continue to rock, with both Seagate and HGST 4TB drives performing well. The Seagate 4TB drive has a current cumulative failure rate of 3.0% and has a street price of $131.58 each on Amazon. The HGST 4TB drive has a higher street price of $174.99 on Amazon, but a lower cumulative failure rate of 1.18%. Both drives have been in service for over a year and we currently own 17,000+ Seagate and 11,000+ HGST 4TB drives and continue to purchase more.

The failure rates of the Toshiba and Western Digital 4TB drives look respectable as well, but they are based on a very limited number of drives for each model. We’ve had trouble getting the Toshiba drives quoted to us in quantity, although there appears to be some movement on that front. Western Digital drives are almost always quoted to us at a higher price than other drive models. Until we can get a reasonable number of these drives, we can’t recommend either of them, although you may find them to be just fine for your personal use.

---

Yev from Backblaze here

It's not a study [...] its just the stats that we have from our datacenter

source: https://www.reddit.com/r/hardware/comments/3ex4aw/backblaze_hard_drive_reliability_stats_for_q2_2015/ctj73dj

---

Seagate still going strong at it that's no laughing matter tho knowing I have a 1Tb SSHDs from them

[Rheinwasser]

i wonder why wd's numbers are worse than hgst's eventhough it's an "wd company."

[Guest]

What about Hitachi?

[Mo5]

Haven't we already established that this is a bunch of BS?

[Centurius]

These guys again?

[zMeul]

What about Hitachi?

Hitachi was acquired by WD in 2012 and the're now known as HGST

[TheMcSame]

What about Hitachi?

Hitachi = HGST these days (though they're two different companies but hitachi drives are from HGST), retailers haven't caught up though so it's not uncommon to see "hitachi" drives.

 

Tests should be taken with a pinch of salt, from what I've heard the conditions are piss poor for hard drives, not to mention the enterprise level workloads.

[ChineseChef]

Haven't we already established that this is a bunch of BS?

 

The problem is the data is accurate, but so many people misuse the data to represent totally different situations.  Then they take this one use case as gospel for why Seagate is just the worst company to exist that is literally run by Hitler. 

 

These aren't made up numbers, but they don't represent 99.9999% of the use cases that regular users will have.

[Centurius]

Hitachi = HGST these days (though they're two different companies but hitachi drives are from HGST), retailers haven't caught up though so it's not uncommon to see "hitachi" drives.

 

Tests should be taken with a pinch of salt, from what I've heard the conditions are piss poor for hard drives, not to mention the enterprise level workloads.

Linus talked about that a year or two back and this article also outlines just how failed their methodology is http://www.tweaktown.com/articles/6028/dispelling-backblaze-s-hdd-reliability-myth-the-real-story-covered/index.html

[Agent181]

These tests are done at high temperature and then put at low temperatures repeatedly then with massive rights and reads so it's at worse case scenario always

[zMeul]

@Centurius

https://www.reddit.com/r/hardware/comments/3ex4aw/backblaze_hard_drive_reliability_stats_for_q2_2015/ctj73dj

[kuddlesworth9419]

I have actually never had a hard drive fail ever. I might only use them for gaming but I think I represent most people when they say HDD's are pretty durable. They might not last as long as an SSD but they work well and are cheap.

[Notional]

My Seagate 7200.11 1TB died too. Actually twice, as I unlocked it with a usb to serial adapter and made it possible for the locked drive to firmware update. Sadly that firmware did nothing good, and the drive died half a year later. Seagate's 7200 series is the worst crap on the earth, as can be seen with even the 7200.14 versions. 12 had major issues too. In another words, Seagate is banned in my mind. Then again, unless I buy a 4TB HDD soon, I will probably never buy an HDD anymore. Don't really need all the space, and SSD's will come in 6 and 8 TB versions soon, which should drive down the price a lot.

[79wjd]

Haven't we already established that this is a bunch of BS?

The problem is the data is accurate, but so many people misuse the data to represent totally different situations.  Then they take this one use case as gospel for why Seagate is just the worst company to exist that is literally run by Hitler. 

 

These aren't made up numbers, but they don't represent 99.9999% of the use cases that regular users will have.

It also kind of raises the question.....what moron at Blackblaze would purchase Seagate HDDs having seen very similar looking graphs year after year. 

[Nineshadow]

BlackBlaze again...

[ChineseChef]

It also kind of raises the question.....what moron at Blackblaze would purchase Seagate HDDs having seen very similar looking graphs year after year. 

 

They keep buying the absolute cheapest drives out there.  Cause they replace in bulk.  Their method doesn't rely on single drive durability.

 

Basically, their data shows that despite the failure rate, it is still more cost effective to buy the "bad" Seagate drives. 

 

Since the Seagate drives are usually considerably cheaper, especially once you start buying drives buy the 100s/1000s.  A $10 or $15 difference per drive is nothing when you are buying a single drive and want them to be reliable for you.  But when you buy 100 drives, that $10 difference means you could buy $1000 worth more of the cheaper drive.

[Ekpyrosis]

What about Samsung/Spinpoint?

[Daegun]

This makes me sad. Not because Backblaze is doing something wrong here (they're just posting pure data, nothing else), but because there are going to be a bunch of idiots that quote this as well as previous released sets to show just how much stronger their argument is for why Seagate drives should never be bought.

 

Idiots are why we can't have nice things.

[spartaman64]

They keep buying the absolute cheapest drives out there.  Cause they replace in bulk.  Their method doesn't rely on single drive durability.

 

Basically, their data shows that despite the failure rate, it is still more cost effective to buy the "bad" Seagate drives. 

 

Since the Seagate drives are usually considerably cheaper, especially once you start buying drives buy the 100s/1000s.  A $10 or $15 difference per drive is nothing when you are buying a single drive and want them to be reliable for you.  But when you buy 100 drives, that $10 difference means you could buy $1000 worth more of the cheaper drive.

but even if they are like 10-15 dollars cheaper i would suppose that what every cost benefit you get is quickly lost and turn to the negatives when 40% of the drives fail

[79wjd]

but even if they are like 10-15 dollars cheaper i would suppose that what every cost benefit you get is quickly lost and turn to the negatives when 40% of the drives fail

If you factor in labor costs and the extra drives required to keep everything running continuously even through large numbers of drive failures then the 7k3000 ($110 according to the article), with a failure rate of around 1.5% would probably be similar to (but still more expensive than) the ST3000MD1 with a failure rate of $30%~ (at around $75). 

[spartaman64]

If you factor in labor costs and the extra drives required to keep everything running continuously even through large numbers of drive failures then the 7k3000 ($110 according to the article), with a failure rate of around 1.5% would probably be similar to (but still more expensive than) the ST3000MD1 with a failure rate of $30%~ (at around $75). 

the grey bar goes to 40%

[ChineseChef]

but even if they are like 10-15 dollars cheaper i would suppose that what every cost benefit you get is quickly lost and turn to the negatives when 40% of the drives fail

 

I agree in principle, you would think having to replace them more often would cost more.  But it is obviously so much cheaper that it offsets any extra costs incurred due to unreliability.

 

At the same time, it doesn't cost much to pay someone barely above minimum wage to swap out bad drives.  And you could even call it a nice entry level position, no 3-5 years experience required, lol.

[spartaman64]

I agree in principle, you would think having to replace them more often would cost more.  But it is obviously so much cheaper that it offsets any extra costs incurred due to unreliability.

 

At the same time, it doesn't cost much to pay someone barely above minimum wage to swap out bad drives.  And you could even call it a nice entry level position, no 3-5 years experience required, lol.

replacing the drive then rebuilding the data is going to lead to a lot of downtime 

[Nineshadow]

45 WD drives vs 17 895 Seagate ones?

 

Jeez

 

Hang on a second...1.5% of their WD drives have failed...that means 0.675 drives failed out of 45.

What...?How can you have 0.6 of a drive to fail?

[ChineseChef]

replacing the drive then rebuilding the data is going to lead to a lot of downtime 

 

Their data systems rebuild automatically, and have extreme redundancy.  They have done some heavy custom programing on their end so that all they have to do is pull out the rack drive drawer, pop out bad drive, pop in new drive, slide system back into rack.  And it will rebuild and recover all on its own, no direct user action required beyond physically swapping the drive.