Wow, Surprised at URE rating on Newer (non-Pro) Barracuda and (non-Pro) Iron Wolf drives

cbn

Lifer
Mar 27, 2009
12,968
221
106
While doing some research on URE ratings for various drives I was quite surprised to find a URE rating of 10^15 (1 per 10E15) for the plain jane Seagate Barracuda drives with SMR (i.e all current drives of this model 4TB and greater as well as some 3TB and 2TB) as well as the non-Pro Iron Worlf drives 6TB and greater. In contrast, the older Seagate Barracuda drives with Conventional magnetic recording (i.e some 3TB and 2TB models + all 1TB and 500GB) as well as the WD Red (non-Pro) have a URE rating of 10^14 (1 per 10E14).

The significance of this, of course, is how much capacity can be used for RAID-5:

https://www.zdnet.com/article/why-raid-5-stops-working-in-2009/

https://www.zdnet.com/article/why-raid-5-still-works-usually/
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
I'd still advise against using RAID5. It's just not that fault tolerant, and very vulnerable during rebuild. Which for larger arrays can take a while (as in days). A URE of 10^15 is still only 125TB worth of writes. We already have 14TB drives, and 16TB have been announced.

Unless you're running some gigantic storage array, there isn't much point beyond RAID1 these days. Then there is the whole data quality issue with that much data.
 
  • Like
Reactions: beginner99

nosirrahx

Senior member
Mar 24, 2018
304
75
101
I agree with that. RAID 1 to mitigate downtime and then caching for performance. RAID 5 with massive drives is just asking for trouble and trouble is what RAID is supposed to help mitigate.
 
  • Like
Reactions: beginner99

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
Using a SMR drive in a RAID array would be idiotic IMO.

They're certainly not ideal for performance RAID. But why would anyone consider using RAID'ed HDDs for anything beyond bulk storage today, when SSDs are so cheap and cheerful? Even the lowliest BX500 will walk over any HDD by an order of magnitude.

As nosirrahx wrote, caching is the way to go.

Hell using one at all isnt a great idea.

They do work well as WORM drives. And that's about it. Frequently changing data or random R/W heavy workload? Forgetaboutit...
 

Rifter

Lifer
Oct 9, 1999
11,522
751
126
They're certainly not ideal for performance RAID. But why would anyone consider using RAID'ed HDDs for anything beyond bulk storage today, when SSDs are so cheap and cheerful? Even the lowliest BX500 will walk over any HDD by an order of magnitude.

As nosirrahx wrote, caching is the way to go.



They do work well as WORM drives. And that's about it. Frequently changing data or random R/W heavy workload? Forgetaboutit...

They would have a purspose for sure if they were alot cheaper than standard drives, like 25% cheaper, but when they are in the same price range as PMR drives which offer consistent performance they really dont have a use case IMO.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Using a SMR drive in a RAID array would be idiotic IMO. Hell using one at all isnt a great idea.

https://www.storagereview.com/seagate_archive_hdd_review_8tb

I did realize that SMR writes slower, but according to that article RAID rebuild times are almost 3x longer with the Seagate Archive SMR HDD. Wow! That is slow!

Still for a 4 x 4TB RAID-5 array what is worse:

1. CMR drives with URE of 1 in 10^14

2. SMR drives with URE of 1 in 10^15

#1 is worse. (reason: even though the rebuild will be faster it has (edit) an almost 97% 96% chance of failing the rebuild. This, compared to the slow writing SMR drives which only have a less than 10% chance of failing the rebuild)

So something to think when deciding on WD (non-Pro) Red and Seagate Barracuda. With that noted, the 7200 rpm (and I believe CMR) non-Pro Iron Wolf (although not available in 4TB with URE 1 in 10^15) could be used as 3 x 6TB in RAID-5 for the same capacity (although it will be $150 more expensive than 4 x 4TB Barracuda).

SIDE NOTE: WD Red Pro 6TB is $206 ---> https://www.newegg.com/Product/Product.aspx?item=N82E16822234344 (only about $22 more than the 6TB (non-Pro) Iron Wolf)


P.S. (Just for record) Seagate Barracuda SMR drives didn't always have URE of 1 in 10^15. I remember seeing 1 in 10^14 last year (I think) and definitely in 2017 the spec was 1 in 10^14. (See the link in this Dec 2017 post I made)
 
Last edited:

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
So something to think when deciding on WD (non-Pro) Red and Seagate Barracuda. With that noted, the 7200 rpm (and I believe CMR) non-Pro Iron Wolf (although not available in 4TB with URE 1 in 10^15) could be used as 3 x 6TB in RAID-5 for the same capacity (although it will be $150 more expensive than 4 x 4TB Barracuda).

Or you could just buy two 12TB drives, and set them in RAID1. Same capacity, but simpler and more reliable.
 
  • Like
Reactions: beginner99

cbn

Lifer
Mar 27, 2009
12,968
221
106
Or you could just buy two 12TB drives, and set them in RAID1. Same capacity, but simpler and more reliable.

That is a very good point you bring up. (RAID 1 and RAID 10 is not subject to URE---> https://community.spiceworks.com/topic/977451-can-raid-10-encounter-a-ure )

But when I looked up pricing I noticed 2 x 12TB in RAID 1 would be ~2 x more expensive (as 4 x 4TB non-Pro Barracuda in RAID 5) and about 1/2 as fast but at the same time more compact and lower power consumption. (Interestingly enough, the workload rate limit of 12TB Iron Wolf and 3 x 4TB Barracuda is not that much different. 180TB for the 12TB Iron Wolf and 3 x 55TB for the 3 x 4TB Barracuda)
 
Last edited:

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
That is a very good point you bring up. (RAID 1 and RAID 10 is not subject to URE---> https://community.spiceworks.com/topic/977451-can-raid-10-encounter-a-ure )

Precisely.

But when I looked up pricing I noticed 2 x 12TB in RAID 1 would be ~2 x more expensive (as 4 x 4TB non-Pro Barracuda in RAID 5) and about 1/2 as fast but at the same time more compact and lower power consumption. (Interestingly enough, the workload rate limit of 12TB Iron Wolf and 3 x 4TB Barracuda is not that much different. 180TB for the 12TB Iron Wolf and 3 x 55TB for the 3 x 4TB Barracuda)

Am I missing something? To get 12TB in a RAID5 array, you'll need 3x 6TB (2 data, 1 parity). 3x 4TB will only net an 8TB capacity.

Even then, for home usage, I'd rather have a single 12TB drive with appropriate backup. If the drive fails, any way you do it, it'll take a while to get data onto a replacement. RAID or no RAID.
 
  • Like
Reactions: beginner99

cbn

Lifer
Mar 27, 2009
12,968
221
106
cbn said:
But when I looked up pricing I noticed 2 x 12TB in RAID 1 would be ~2 x more expensive (as 4 x 4TB non-Pro Barracuda in RAID 5) and about 1/2 as fast but at the same time more compact and lower power consumption. (Interestingly enough, the workload rate limit of 12TB Iron Wolf and 3 x 4TB Barracuda is not that much different. 180TB for the 12TB Iron Wolf and 3 x 55TB for the 3 x 4TB Barracuda)

Am I missing something? To get 12TB in a RAID5 array, you'll need 3x 6TB (2 data, 1 parity). 3x 4TB will only net an 8TB capacity.

I apologize for not clarifying.

I left out the mirrored drive in the RAID-1 volume and parity drive in the RAID-5 volume to determine the workload rate limit of the volumes.

2 x 12TB in RAID-1 and each 12TB drive has a workload rate limit of 180 then workload rate limit of the volume is 180.

4 x 4TB RAID-5 and each 4TB drive has a workload rate limit of 55 then workload rate limit of the volume is 165.
 
  • Like
Reactions: Insert_Nickname

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
I apologize for not clarifying.

I left out the mirrored drive in the RAID-1 volume and parity drive in the RAID-5 volume to determine the workload rate limit of the volumes.

2 x 12TB in RAID-1 and each 12TB drive has a workload rate limit of 180 then workload rate limit of the volume is 180.

4 x 4TB RAID-5 and each 4TB drive has a workload rate limit of 55 then workload rate limit of the volume is 165.

Workload ratings are not cumulative. All data and associated parity has to be written to all drives. So even if an array consists of 4 drives, the workload rating is still only 55TB. RAID5 is further susceptible to write amplification.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Workload ratings are not cumulative. All data and associated parity has to be written to all drives. So even if an array consists of 4 drives, the workload rating is still only 55TB. RAID5 is further susceptible to write amplification.

It should be cumulative because any data written is being split up among more than one drive.

For example, If 165TB is written to a RAID-5 volume of 4 x 4TB each drive only gets 55TB.
 
Last edited:

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
It should be cumulative because any data written is being split up among more than one drive.

For example, If 165TB is written to a RAID-5 volume of 4 x 4TB each drive only gets 55TB.

I think we're talking past each other. It might have been poorly worded on my part.

Per drive workload isn't cumulative, but the array can handle more data written then the individual drives themselves. Because data and parity is striped across multiple drives. But the individual drives will still have a 55TB limit. You just can't put f.x. 4 drives together and suddenly get a 220TB per drive workload rating. But a RAID5 array of those drives would have that workload rating.

I actually think we're talking about the same thing, from different perspectives.

I hope that makes sense.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
You just can't put f.x. 4 drives together and suddenly get a 220TB per drive workload rating. But a RAID5 array of those drives would have that workload rating.

The work rate limit of a RAID-5 array of 4 x 4TB drives (each with work rate limit of 55) would be 165.

Likewise the work rate limit of a RAID-1 array of 2 x 12TB drives (each with a work rate limit of 180) would only be 180, not 360.
 
Last edited:

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
The work rate limit of a RAID-5 array of 4 x 4TB drives (each with work rate limit of 55) would be 165.

You're forgetting parity. Data is written across 3 drives, with parity on the 4th. So you're still writing to all drives in the array with every write. So it would be 220TB for the array.

Likewise the work rate limit of a RAID-1 array of 2 x 12TB drives (each with a work rate limit of 180) would only be 180, not 360.

Yup. RAID1 does writes in parallel across each drive. So it's limited to a single drive rating.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
You're forgetting parity. Data is written across 3 drives, with parity on the 4th. So you're still writing to all drives in the array with every write.

Yes, I totally understand that. This is why I wrote the following in post #13:

cbn said:
For example, If 165TB is written to a RAID-5 volume of 4 x 4TB each drive only gets 55TB.


So it would be 220TB for the array.

165.

.....just like a 4 x 4TB (ST4000DM004) RAID-5 volume is only 12TB (not 16TB), and only 3 x 190 MB/s Sequential Read (not 4 x 190 MB/s Sequential Read) , work rate limit is only 165 (not 220).
 
Last edited:

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
165.

.....just like a 4 x 4TB (ST4000DM004) RAID-5 volume is only 12TB (not 16TB), and only 3 x 190 MB/s Sequential Read (not 4 x 190 MB/s Sequential Read) , work rate limit is only 165 (not 220).

Parity is generated from the data you're writing, then written across each disk. Effectively the array is 16TB. The usable capacity is 12TB.

I see no point in continuing this.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
The work rate limit of a RAID-5 array of 4 x 4TB drives (each with work rate limit of 55) would be 165.

You're forgetting parity. Data is written across 3 drives, with parity on the 4th. So you're still writing to all drives in the array with every write. So it would be 220TB for the array.

165.

.....just like a 4 x 4TB (ST4000DM004) RAID-5 volume is only 12TB (not 16TB), and only 3 x 190 MB/s Sequential Read (not 4 x 190 MB/s Sequential Read) , work rate limit is only 165 (not 220).

Parity is generated from the data you're writing, then written across each disk. Effectively the array is 16TB. The usable capacity is 12TB.

The only way a RAID array of four 4TB disks (each with a work rate limit of 55) could handle 220TB data in one year is if they were in RAID-0.

220TB of data through four 4TB disks (each with a work rate limit of 55) in RAID-5 exceeds the work rate limit of the array. (and if a drive broke at this point it would be out of warranty because it has 73TB of use. 18TB more than the warranty limit) The only way to stay within the warranty limit of four 4TB drives (each with work rate limit of 55) in RAID-5 is to limit usage to 165TB and below.
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
According to Seagate spec sheets the Enterprise 10,000 rpm and 15,000 rpm drives have a URE spec of 1 in 10^16.

I wonder how long it will take for the capability to migrate to lower end drives? Maybe "lower end" multi-actuator drives? (There was a time I thought I would never see URE of 1 in 10^15 on a cheap drive)
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Even then, for home usage, I'd rather have a single 12TB drive with appropriate backup. If the drive fails, any way you do it, it'll take a while to get data onto a replacement. RAID or no RAID.

Excatly. RAID is for up-time which is irrelevant for home-use. As a home user, RAID is a waste of money. Just work with single drives and make regular backups to an external drive which you store somewhere else (easiest being workplace).
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
As a home user, RAID is a waste of money. Just work with single drives and make regular backups to an external drive which you store somewhere else (easiest being workplace).

Single drives don't have enough sequential for 4K editing with an intermediate codec.

But with a cheap RAID-5 (or storage spaces volume) a person could do that.

I personally like the idea of a basic build with a C drive made up with this RAID-5 using Optane for cache. (Just back-up essential files to an external drive)
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
I personally like the idea of a basic build with a C drive made up with this RAID-5 using Optane for cache. (Just back-up essential files to an external drive)

That sounds overly complex for now apparent gain. Just get an ssd but I doubt that a fast spiner doesn't have neough bandwidth, fastes on easily reach >150 mb/s now.
 
  • Like
Reactions: Insert_Nickname

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
That sounds overly complex for now apparent gain. Just get an ssd but I doubt that a fast spiner doesn't have neough bandwidth, fastes on easily reach >150 mb/s now.

Ah, but for random 4K R/W you really shouldn't expect more then 1.5-2MB/s. I shudder to think how long it'd take to fill even a 4TB drive at that speed. A month or two?

Lets see. 4TB = 4000GB = 4.000.000MB (yes, yes, you should subtract 7%. Just indulge me)

So it'd take 4.000.000 seconds to fill @ 1MB/s. That's 66666.7 minutes, 1111 hours or about 46 days. Ouch...
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
(snip) I doubt that a fast spiner doesn't have neough bandwidth, fastes on easily reach >150 mb/s now.

Here is a quote from a B and H Photo article:

https://www.bhphotovideo.com/explor...ns/hard-drive-solutions-video-editing-studios

Once you know how much bandwidth a single track of the codec requires, your next step is to determine how many tracks you are going to need to play back at once. If you will only be doing simple editing with some transitions and text overlays, then twice the bandwidth is probably enough. However, if you plan on doing multi-camera cutting in real time, or using a large number of layers simultaneously, you are going to need far more.

And some data rates from Avid:

http://avid.force.com/pkb/articles/en_US/White_Paper/DNxHR-Codec-Bandwidth-Specifications

Screenshot-56.png


Even simple 4K editing with some transitions and text overlays using DNxHR HQX at 4K30 is going to require 222.2 MB/s. (re: 2 x 111.10 = 222.2 MB/s). Same, but at 4K60, would require 444.4 MB/s (re: 2 x 222.2 MB/s).
 
Last edited: