Backblaze - "Consumer hard drives as reliable as enterprise hardware"

[jumperalex]

For the first three years at least.  After that they can't say ... yet.

 

http://www.pcpro.co.uk/news/385792/consumer-hard-drives-as-reliable-as-enterprise-hardware

[wsume99]

That's funny. A few weeks ago I was going back and forth in another thread with a moderator on this forum about this exact topic. He claimed that the higher class drives (ie NAS or Enterprise) were more reliable than desktop class drives. I did not agree and asked for data to backup such an assertion. He provided MTBF ratings from the OEM as evidence. The kind of life analysis performed by Backblaze outlined in this article is more along the lines of what I was thinking and IMO pretty much proves my point.

[archedraft]

Defiantly an interesting read! I will look forward to reading if there is a longer life period of the enterprise drives. Although, if the median for consumer drives is 6 years, I am sure for me that after that time I will be thinking wow I still have this tiny 3TB hard drive? Time to upgrade to the 50 TB drive... lol.

 

On a side note, I wonder if Blackblaze has any sort of bias towards consumer drives. I know absolutely nothing about Blackblaze but would they have anything to gain from claiming that consumer drives are similar to enterprise drives. It would make sense to me that whomever choose to use the consumer drives in Blackblaze would be wanting these results (it makes that person(s) look good). Not trying to stir the pot but I always like to think about what / why a company would want to make the their data available to the public, even if the goal was to just help others.

[jumperalex]

Exactly.  There has rarely ever been evidence in situations like this that the products are actually produced any differently.  The company just does the actuarial analysis to determine how much extra to charge to cover the extra years of warranty based on expected return rate.  Which is the same analysis they do for the shorter warranty period.  It would just cost too much to run two lines, two levels of tolerance control, two non-linked QA tests, or even two levels of component sourcing.  They would lose economies of scale and savings from consolidation. 

 

The best you could hope for is to bin devices based on something they know as a normal course of production / testing that is a strong predictor for reliability (read: they already measure it for everything).  I don't know what that would be for an HDD.  Maybe something like measured platter flatness, or bearing race tolerance after machining [shrug]

 

Its the same idea as they do with CPUs. Other than actual architectural design differences, they build them based on an expect range of yield, test them for stability up to certain freqs at certain voltage, and bin them.  In this case the QA is linked because they can non-destructively test to failure but still qualify for a lower QA level.  Heck they even sell 2/3/6 core chips which are nothing my than 4/8 core chips with 1/2 failed cores.  They've even been known to sell chips with less cache memory because their yield for that part of the die wasn't as good as they expected, rather than spend the money trying to redesign the die / improve the process.

[jumperalex]

On a side note, I wonder if Blackblaze has any sort of bias towards consumer drives. I know absolutely nothing about Blackblaze but would they have anything to gain from claiming that consumer drives are similar to enterprise drives. It would make sense to me that whomever choose to use the consumer drives in Blackblaze would be wanting these results (it makes that person(s) look good). Not trying to stir the pot but I always like to think about what / why a company would want to make the their data available to the public, even if the goal was to just help others.

 

An interesting question.  I'd be more inclined to want to see the results if someone talked me into the more expensive enterprise drives.  It surely wouldn't help on the supply/demand side of things as demand for consumer drives rises.  And their is a premium they can charge for enterprise drives which tends to increase profit margin, so that seems counter productive.  But I certainly could see them wanting to show "they were right" to the customers they talked into consumer drives.  Call it good-will-equity.  Similarly it makes them look good to potential consumers who think, "hey those guys are straight shooters".  But boy they better be telling the truth though because that type of things bites you in the butt if shown to be false.  And "going public" as a form of advertising would work even better with enterprise drives because now they can charge a premium for the drives AND get the good-will-equity of showing proof.

[BobPhoenix]

I can say I had HORRIBLE luck with WD Green drives and I don't consider the Red's to be an enterprise drive.  The Reds are a consumer drive with a NAS firmware and longer warranty.  A true enterprise drive has components that allow the NREs to be better by an order of 10.  NREs with an enterprise drive are <10 in 10 to the 15th power bits read and consumer/Red drives are rated at <10 in 10 to the 14th bits read.  Now whether that really makes a difference in real use or not I can't say and the Backblaze results seem to indicate it DOESN'T matter.  So as I said in that "other" thread for me the longer warranty and (so far at least) better personal experience with the Reds make them worth the added cost.

 

 

I had 28 2&3TB green drives. 12 are now showing smart errors but would otherwise be in good shape.  Another 12 I toasted with a 60+C parity check and now have smart errors as well.  Out of the 28 drives I only have 4 2TB greens that are still in acceptable working condition.  One of the 3TB drives might be fine but it flips between 65535 pending and zero pending depending on the preclear cycle - it was suggested it was a firmware problem to me by other forum members

[jumperalex]

Yeah Red bring other things to the table, mostly firmware. 

 

As I'm ignorant of the details of NRE (I know it is non-recoverable errors at least) I ask you: is that something that is more firmware related or hardware related?  Both?  Would it be effected by tracking accuracy or platter coating depth / uniformity?  Given the same basic hardware and firmware algorithms (read not using faster chips) can you reduce NRE by reducing performance?  Even just marginally?

 

Those are the things that come to mind (literally just guessing) that could be specifically engineered / selected for to assure higher reliability.

[BobPhoenix]

Have no clue.  That is just one reason I've seen posted elsewhere why the enterprise drives are better.  For me the cheaper cost of the desktop models and the frequency of replacing them to get bigger models means I'm ok with consumer drives with the exception being a WD Green - not getting one again due to the high failure rate I got.  The NAS (Red) drives (so far anyway) seem to be working better for me.  Note I have not bought a Seagate NAS model yet just have DM models and earlier.  But I came close on Friday.  Just wanted to wait for Monday to see if any other drives would be cheaper and the Seagate NAS were not on sale Monday so went to more DMs since they WERE on sale on Monday.

[lionelhutz]

The guys at Backblaze seem to tell it like it is. They have thrown out some interesting information over the years, such as how they build their cheap server boxes and how they were finding the Hitachi drives more reliable then WD or Seagate back before the floods.

[Fireball3]

What about this?

 

He also pointed out that the company's usage of the drives is different, with enterprise drives being used more heavily than their consumer counterparts.

 

Comparing apples and oranges???

 

[jumperalex]

You are right .. but keep reading.  He also says they are more pampered too.

 

However, he noted that the enterprise drives are "coddled in well-ventilated, low-vibration enclosures".

 

Read more: Consumer hard drives as reliable as enterprise hardware | News | PC Pro http://www.pcpro.co.uk/news/385792/consumer-hard-drives-as-reliable-as-enterprise-hardware#ixzz2mc3fy23t

 

so which is more important?  Got me.  But regardless what they are seeing across a lot of drives, and drives that are likely still being hammered way more than any home user / unraid user will, is very little difference in failure rates.  That puts the onus of proof on the enterprise marketing department to prove their case.  Or should I say puts it more firmly since they already had a burden of proof that has never been satisfied beyond their own quoted, and mostly debunked, superior MTBF rates and I suspected even the previously discussed NRE rates.

[wsume99]

What about this?

 

He also pointed out that the company's usage of the drives is different, with enterprise drives being used more heavily than their consumer counterparts.

 

Comparing apples and oranges???

 

Well why not include both sentences from that paragraph ...

He also pointed out that the company's usage of the drives is different, with enterprise drives being used more heavily than their consumer counterparts. However, he noted that the enterprise drives are "coddled in well-ventilated, low-vibration enclosures".

So it is possible that the enterprise drives are in a better operating environment than the consumer drives. So maybe it's a wash with the heavier usage. Who knows. What we do know is that from their data there is not a significant difference in drive failure rate - at least nothing even close to approaching the difference in reliability that the OEM's claim in their specs.

[jumperalex]

Dude you are my conversational doppelganger!!!

[Fireball3]

<blockquote>

He also pointed out that the company's usage of the drives is different, with enterprise drives being used more heavily than their consumer counterparts. However, he noted that the enterprise drives are "coddled in well-ventilated, low-vibration enclosures".

</blockquote>

So it is possible that the enterprise drives are in a better operating environment than the consumer drives. So maybe it's a wash with the heavier usage. Who knows. What we do know is that from their data there is not a significant difference in drive failure rate - at least nothing even close to approaching the difference in reliability that the OEM's claim in their specs.

 

While enterprise drives produce considerably more heat it is necessary to cool them really good.

Asuming they are at the same temperature level it is still not the same load they are working at.

 

If testing is done at different load levels you can't compare reliability figures.

[wsume99]

<blockquote>

He also pointed out that the company's usage of the drives is different, with enterprise drives being used more heavily than their consumer counterparts. However, he noted that the enterprise drives are "coddled in well-ventilated, low-vibration enclosures".

</blockquote>

So it is possible that the enterprise drives are in a better operating environment than the consumer drives. So maybe it's a wash with the heavier usage. Who knows. What we do know is that from their data there is not a significant difference in drive failure rate - at least nothing even close to approaching the difference in reliability that the OEM's claim in their specs.

 

While enterprise drives produce considerably more heat it is necessary to cool them really good.

Asuming they are at the same temperature level it is still not the same load they are working at.

 

If testing is done at different load levels you can't compare reliability figures.

We don't know exactly how the usage compares so at this point all we can do is believe them or not. Right now their money is betting on having a lower total cost of ownership from desktop class HDDs. Seems to me that Backblaze is running their own little experiment to make sure that they are using the most economical setup. At this point they seem to feel confident that their results have proven them correct.

[c3]

BackBlaze only puts consumer data on consumer drives.

 

The one thing the google study did show, usage/load did not change drive failure rates. Temp maybe, age bathtub curve

[NAS]

I have been waiting for this specific blog post as it has been promised for a while.

 

My take on this. First off I have learned to trust Backblaze. As posted earlier they have a 100% history of "telling it how it is" and a surprisingly open culture for a purely commercial company. Is it possible they have a ulterior motive or are simply wrong, sure, but personally i don't think so.

 

The results they posted don't surprise me. My experience has never been that enterprise class drives were more reliable. Admittedly I haven't purchased as many (about 100 this year) so i speak with a little experience.

 

However whilst it is easy to pick comparative and statistical holes in their results and conclusions I do not think that is the point.

 

The real question is why people are keen to prove their work is wrong head rather than provide info that proving otherwise.

 

As far as I know there no there no quantitative open analysis to prove the long standing myth that enterprise class drives are THAT much better (or in fact any better at all). Sure they have a higher MTBF (which is meaningless in the context of small numbers of drives) a higher warranty (but you pay for it) and refined characteristics for certain workloads (I would argue that the days of this being enough to make it worth the money are gone) but is that enough to justify the cost?

 

[dgaschk]

This validates Googles study: http://static.googleusercontent.com/media/research.google.com/en/us/archive/disk_failures.pdf

Large data companies, e.g. Backblaze and Google,  are not using enterprise drives because the TCO does not support their use. Errors are handled by higher level processes, i.e. file checksums and redundancy, so individual drive errors and failures do not effect data integrity and availability.