Tuesday, April 30th 2019

Backblaze Releases Hard Drive Stats for Q1 2019

As of March 31, 2019, Backblaze had 106,238 spinning hard drives in our cloud storage ecosystem spread across three data centers. Of that number, there were 1,913 boot drives and 104,325 data drives. This review looks at the Q1 2019 and lifetime hard drive failure rates of the data drive models currently in operation in our data centers and provides a handful of insights and observations along the way.

Hard Drive Failure Stats for Q1 2019
At the end of Q1 2019, Backblaze was using 104,325 hard drives to store data. For our evaluation we remove from consideration those drives that were used for testing purposes and those drive models for which we did not have at least 45 drives (see why below). This leaves us with 104,130 hard drives. The table below covers what happened in Q1 2019.
Notes and Observations
If a drive model has a failure rate of 0%, it means there were no drive failures of that model during Q1 2019. The two drives listed with zero failures in Q1 were the 4 TB and 5 TB Toshiba models. Neither has a large enough number of drive days to be statistically significant, but in the case of the 5 TB model, you have to go back to Q2 2016 to find the last drive failure we had of that model.

There were 195 drives (104,325 minus 104,130) that were not included in the list above because they were used as testing drives or we did not have at least 45 of a given drive model. We use 45 drives of the same model as the minimum number when we report quarterly, yearly, and lifetime drive statistics. The use of 45 drives is historical in nature as that was the number of drives in our original Storage Pods. Beginning next quarter that threshold will change; we'll get to that shortly.

The Annualized Failure Rate (AFR) for Q1 is 1.56%. That's as high as the quarterly rate has been since Q4 2017 and its part of an overall upward trend we've seen in the quarterly failure rates over the last few quarters. Let's take a closer look.
Quarterly Trends

We noted in previous reports that using the quarterly reports is useful in spotting trends about a particular drive or even a manufacturer. Still, you need to have enough data (drive count and drive days) in each observed period (quarter) to make any analysis valid. To that end the chart below uses quarterly data from Seagate and HGST drives while leaving out Toshiba and WDC drives as we don't have enough drives from those manufacturers over the course of the last three years.
Over the last three years, the trend for both Seagate and HGST annualized failure rates had improved, i.e. gone down. While Seagate has reduced their failure rate over 50% during that time, the upward trend over the last three quarters requires some consideration. We'll take a look at this and let you know if we find anything interesting in a future post.
Changing the Qualification Threshold

As reported over the last several quarters, we've been migrating from lower density drives, 2, 3, and 4 TB drives, to larger 10, 12, and 14 TB hard drives. At the same time, we have been replacing our stand-alone 45-drive Storage Pods with 60-drive Storage Pods arranged into the Backblaze Vault configuration of 20 Storage Pods per vault. In Q1, the last stand-alone 45-drive Storage Pod was retired. Therefore, using 45 drives as the threshold for qualification to our quarterly report seems antiquated. This is a good time to switch to using Drive Days as the qualification criteria. In reviewing our data, we have decided to use 5,000 Drive Days as the threshold going forward. The exception, any current drives we are reporting, such as the Toshiba 5 TB model with about 4,000 hours each quarter, will continue to be included in our Hard Drive Stats reports.
Fewer Drives = More Data

Those of you who follow our quarterly reports might have observed that the total number of hard drives in service decreased in Q1 by 648 drives compared to Q4 2018, yet we added nearly 60 petabytes of storage. You can see what changed in the chart below.
Lifetime Hard Drive Stats
The table below shows the lifetime failure rates for the hard drive models we had in service as of March 31, 2019. This is over the period beginning in April 2013 and ending March 31, 2019.
Source: Backblaze
Add your own comment

30 Comments on Backblaze Releases Hard Drive Stats for Q1 2019

#1
dj-electric
"lets take a bunch HDDs, use them for server use, and let the whole internet catch on fire again"
Posted on Reply
#2
R-T-B
dj-electric, post: 4039502, member: 87186"
"lets take a bunch desktop HDDs, use them for server use, and let the whole internet catch on fire again"
Missed the most important detail...

if they used server HDDs the data MIGHT be vaguely useful.
Posted on Reply
#3
natr0n
I enjoy these lists.

Seagate is the winner of the losers.
Posted on Reply
#4
EarthDog
The take away again...any of them will be fine for a consumer with such low failure rates. If you are splitting hairs between 2 of every 100 or 1 of every hundred, you need physical redundancy and a good backup plan (as always) not a HDD with an ever so slightly improved failure rate (in a data center which is an unrealistic use model for consumer drives anyway).
Posted on Reply
#5
Fouquin
R-T-B, post: 4039508, member: 41983"
if they used server HDDs the data MIGHT be vaguely useful.
They did use server HDDs.

Toshiba MG07ACA14TA - 9-disk He-Sealed 7200RPM Enterprise HDD
HGST HUH721212ALE600/ALN604 - HC520 He-Sealed 7200RPM Enterprise HDD
Seagate ST12000NM0007 - Exos x12 He-Sealed 7200RPM Enterprise HDD
Seagate ST10000NM0086 - Exos x10 He-Sealed 7200RPM Enterprise HDD

I think you get the picture. The lower capacity Seagate drives are desktop class and have been used continually by Backblaze for a few years. They offer a consumer baseline to compare against. The other vendor's drives are long-service NAS or Surveillance drives, which some might argue should have a similar life span to an Enterprise drive.
Posted on Reply
#6
PerfectWave
i dont think those data are useful tbh. you dont know what kind of workload the hd was used
Posted on Reply
#7
yakk
This report makes WD look good, just because they're practically not included in there.

Funny that.
Posted on Reply
#8
Ahhzz
They missed where they took out 180 WD 3Tb drives.... Did I miss where they explained that?

Ah, nvm. Their chart is from Q4 2018. Looks like they pulled them from service somewhere last year, and wouldn't have explained it again. Thought it was Q1 to Q1.
Posted on Reply
#9
R-T-B
Fouquin, post: 4039527, member: 157604"
The other vendor's drives are long-service NAS or Surveillance drives, which some might argue should have a similar life span to an Enterprise drive.
I wouldn't go that far, but they are a good middleground.

They certainly are at least trying this time around to address some longstanding concerns: Good. They still have some major statistical representation issues, but at least it's more useful than nothing now.
Posted on Reply
#10
Juventas
Yes, they really should just give the names in the chart, with models in the article.

HMS5C4040ALE640 MegaScale DC
HMS5C4040BLE640 MegaScale DC
HUH728080ALE600 Ultrastar He8
HUH721212ALN604 Ultrastar DC HC520 (aka He12)
ST4000DM000 Desktop HDD
ST6000DX000 Desktop HDD
ST8000DM002 Desktop HDD
ST8000NM0055 Exos 7E8
ST10000NM0086 Exos X10
ST12000NM0007 Exos X12
MD04ABA400V Surveillance HDD
MD04ABA500V Surveillance HDD
MG07ACA14TA Enterprise HDD
WD60EFRX Red NAS

So, not really a fair comparison. HGST is all enterprise drives. Seagate and Toshiba is a mix. WD is for SOHO NAS (i.e. they're not Red Pro).
Posted on Reply
#11
Chloe Price
Aaaand Seagate leads the stats. What a surprise..
Posted on Reply
#12
R-T-B
Chloe Price, post: 4039643, member: 123719"
Aaaand Seagate leads the stats. What a surprise..
I like how everyone auto-assumes that and misses the fact the worst drive in this list by AFR is a HGST...
Posted on Reply
#13
Chloe Price
R-T-B, post: 4039779, member: 41983"
I like how everyone auto-assumes that and misses the fact the worst drive in this list by AFR is a HGST...
Heey hold on a minute, I have a Seagate drive on my PS2 and it works..


These stats just mean that I wouldn'd be buying these or recommending to my friends.
Posted on Reply
#14
EarthDog
R-T-B, post: 4039779, member: 41983"
I like how everyone auto-assumes that and misses the fact the worst drive in this list by AFR is a HGST...
For consumer drives and more realistic use scenarios, I would look at Puget Systems's data. Again, I think we'll find that outside of a 'bad' drive most are not worth fretting over the % failure. I get that one drive might have 2x the failure rate of another, but, that is the difference between 99/100 working or 99.5/100 working. Context is important here. :)

https://www.pugetsystems.com/labs/articles/Most-Reliable-PC-Hardware-of-2018-1322/#Storage(HardDrive)
Posted on Reply
#15
pjl321
Will they ever move over to SSDs or is the improved failure rate and performance of an SSD totally insignificant to them in comparison to the cost of HDDs?
Posted on Reply
#16
EarthDog
I used to work for AWS and it depends on the system as to what drives are used. If speed is needed over capacity, SSDs are used. Otherwise, its platter storage.

That said, AWS isn't really a backup joint, primarily, so there's that.

But yes, I would imagine it is cost prohibitive still to run all high capacity SSDs. It takes almost no time for a tech to replace a platter/HDD and get the array rebuilt, versus the HUGE cost difference between comparable capacity platters/SSD.
Posted on Reply
#17
pjl321
EarthDog, post: 4039940, member: 79836"
I used to work for AWS and it depends on the system as to what drives are used. If speed is needed over capacity, SSDs are used. Otherwise, its platter storage.

That said, AWS isn't really a backup joint, primarily, so there's that.

But yes, I would imagine it is cost prohibitive still to run all high capacity SSDs. It takes almost no time for a tech to replace a platter/HDD and get the array rebuilt, versus the HUGE cost difference between comparable capacity platters/SSD.
I had a look around for SSD failure rate figures when choosing what drives to buy but couldn't really find anything reliable like these mass results.
Posted on Reply
#18
EarthDog
AWS doesn't post their failure rates...but I can tell you daily technicians replaced 5-10 drives /day of thousands.

The best you have may be Puget Systems I referenced earlier.
Posted on Reply
#19
Chloe Price
pjl321, post: 4039935, member: 70886"
Will they ever move over to SSDs or is the improved failure rate and performance of an SSD totally insignificant to them in comparison to the cost of HDDs?
For servers etc. SSD's aren't just yet the thing.


I love WD drives myself, damn that Samsung sold its HDD business. They were also awesome.
Posted on Reply
#20
Ubersonic
pjl321, post: 4039935, member: 70886"
Will they ever move over to SSDs or is the improved failure rate and performance of an SSD totally insignificant to them in comparison to the cost of HDDs?
SSDs have been used in high end file servers for years (enterprise SSDs not consumer SSDs obviously), it's only really huge archive servers or budget small business servers that still use HDDs.
Posted on Reply
#21
EarthDog
Chloe Price, post: 4040059, member: 123719"
For servers etc. SSD's aren't just yet the thing.
They can be though. It just depends on the use. At least at AWS, and I would assume with Aszure, it depends on what type of server you need as to what the storage is like. Hell, I know we had ROWS, like 20 yards long with complete SSD systems for some of their services.

Some even have a small SSD for the OS to boot from (If it isn't PXE booted and an image dumped on it).
Posted on Reply
#22
Gasaraki
dj-electric, post: 4039502, member: 87186"
"lets take a bunch HDDs, use them for server use, and let the whole internet catch on fire again"
What does that even mean?

PerfectWave, post: 4039557, member: 165997"
i dont think those data are useful tbh. you dont know what kind of workload the hd was used
Hard drive in RAID arrays have equal usage.
Posted on Reply
#23
Chloe Price
EarthDog, post: 4040094, member: 79836"
They can be though. It just depends on the use. At least at AWS, and I would assume with Aszure, it depends on what type of server you need as to what the storage is like. Hell, I know we had ROWS, like 20 yards long with complete SSD systems for some of their services.

Some even have a small SSD for the OS to boot from (If it isn't PXE booted and an image dumped on it).
Yeah I guess. But for servers, they should be some industrial shit, just like SCSI/SAS drives.

I know too many people who think that you can just use consumer drives in servers, well, you can, but.... They exist for a reason. Server hardware, I mean.
Posted on Reply
#24
newtekie1
Semi-Retired Folder
pjl321, post: 4039935, member: 70886"
Will they ever move over to SSDs or is the improved failure rate and performance of an SSD totally insignificant to them in comparison to the cost of HDDs?
They only care about cost/GB, speed of the drives doesn't matter to them.
Posted on Reply
#25
EarthDog
Chloe Price, post: 4040147, member: 123719"
Yeah I guess. But for servers, they should be some industrial shit, just like SCSI/SAS drives.

I know too many people who think that you can just use consumer drives in servers, well, you can, but.... They exist for a reason. Server hardware, I mean.
Of course they are enterprise SSDs...

You can easily use consumer grade hardware in servers. There isn't a terrible amount of difference in a lot of the hardware. Obviously enough for it to be 'worth it' to the enterprise, but, yeah, they in data centers and a large part of them. :)
Posted on Reply
Add your own comment