Building a cold storage server with Seagate Archive

[zonderling]

Linus explains it better then myself, might be you guys find it interesting.

 

 

[JorgeB]

I only objected to your "will fail fast" statement, mtbf doesn't really mean much to me, e.g., WD stopped using that (https://support.wdc.com/knowledgebase/answer.aspx?ID=665)

 

I don't dispute the WD RED is probably more reliable, although we'll still need to wait some time to see if there are issues with helium leaking , but IMO for the price, the Seagate Archive is a very good option for unRAID, including for parity.

 

If the user is going to be making simultaneous writes to more than one data disk and/or work with a lot of small files, then I would agree that a non SMR drive would be a better choice.

[garycase]

For the use outlined here, the Seagate archive drives are just fine -- including for parity.   They do NOT have any write penalties UNLESS you "hit the wall" where the persistent cache is filled (which will dramatically slow things down until the cache is emptied).    For an array where you (a) aren't writing to multiple disks are once, and (b) are changing relatively little data;  the archive drives perform every bit as good as a non-SMR WD Red (for example).

 

As for single vs. dual parity => as already outlined, it's simply a matter of whether or not you want to risk a failed rebuild due to a 2nd drive failure during the rebuild operation.   Dual parity protects against that.    Doesn't matter how many disks you have -- 6 disks or 26 disks, you could still have a 2nd failure.   Clearly the more disks you have the higher the likelihood of the 2nd failure -- but if you're not backed up (as you indicated is the case) and want to absolutely minimize the likelihood of data loss due to this scenario, a 2nd parity drive isn't a bad bit of extra "insurance".

 

FWIW, if you DO want an 8TB WD Red, and don't mind "shucking" the drive from an external unit, there's a very good deal at the moment where these are actually cheaper than the Seagate archives:   Best Buy has the 8TB "EasyStore" drives for $180 ... these have 8TB Reds inside, which you can remove and use in an array.   Note that there are warranty implications of doing this -- you'd likely need to replace the drive in the case if you ever needed warranty service (not hard to do as long as you're CAREFUL and don't damage the case when removing the drive); and the warranty is 2 years vs. the 3 years that a bare drive comes with.   Nevertheless, it's not a bad option:  http://www.bestbuy.com/site/wd-easystore-8tb-external-usb-3-0-hard-drive-black/5792401.p?skuId=5792401

 

[volume]

I only objected to your "will fail fast" statement, mtbf doesn't really mean much to me, e.g., WD stopped using that (https://support.wdc.com/knowledgebase/answer.aspx?ID=665)
 
I don't dispute the WD RED is probably more reliable, although we'll still need to wait some time to see if there are issues with helium leaking , but IMO for the price, the Seagate Archive is a very good option for unRAID, including for parity.
 
If the user is going to be making simultaneous writes to more than one data disk and/or work with a lot of small files, then I would agree that a non SMR drive would be a better choice.

No there are no need for simultaneous writes. I will only move my data for xpenology to unraid. So I think the seagate archive is a good choice.

[zonderling]

22 minutes ago, johnnie.black said:

I only objected to your "will fail fast" statement, mtbf doesn't really mean much to me, e.g., WD stopped using that (https://support.wdc.com/knowledgebase/answer.aspx?ID=665)

 

I don't dispute the WD RED is probably more reliable, although we'll still need to wait some time to see if there are issues with helium leaking , but IMO for the price, the Seagate Archive is a very good option for unRAID, including for parity.

 

If the user is going to be making simultaneous writes to more than one data disk and/or work with a lot of small files, then I would agree that a non SMR drive would be a better choice.

I'm sorry, but I feel you are neglecting the context of my statement.  I think they "will fail fast" as parity disk.  (the last context is essential ) just because it will sustain a lot of small writes each time something is written on the array.

 

For general purpose, I like them a lot, don't get me wrong.  Best bang for the buck money wise.  

[JorgeB]

8 minutes ago, zonderling said:

I think they "will fail fast" as parity disk.  (the last context is essential )

 

Doesn't change the fact that this is your "feeling", no evidence of this whatsoever.

[garycase]

4 minutes ago, zonderling said:

I think they "will fail fast" as parity disk.  (the last context is essential ) just because it will sustain a lot of small writes each time something is written on the array.

 

Parity or not, there's no issue with these drives as parity drives as long as you don't have activity that fills the persistent cache -- and Seagate has done an excellent job of mitigating this impact, so it's VERY unlikely this will ever happen in most UnRAID scenarios -- especially the use case outlined here.

 

There's a fairly extensive discussion on these drives here: 

Note particularly the extensive experience that Danioj has had with these drives.

 

[garycase]

1 minute ago, johnnie.black said:

no evidence of this whatsoever.

 

Agree.

 

[zonderling]

5 minutes ago, johnnie.black said:

 

Doesn't change the fact that this is your "feeling", no evidence of this whatsoever.

My "feeling" is usually right lol  but Ok; agreed, no numbers to stand me by, only common sense because mechanical ware and tear because of the way The shingled technology works. More movements of the head an all ...

 

But the more important thing remains:  performance issue on writes.  I always want my most performant spinner as parity to avoid bottleneck.

 

[zonderling]

40 minutes ago, garycase said:

FWIW, if you DO want an 8TB WD Red, and don't mind "shucking" the drive from an external unit, there's a very good deal at the moment where these are actually cheaper than the Seagate archives:   Best Buy has the 8TB "EasyStore" drives for $180 ... these have 8TB Reds inside, which you can remove and use in an array.   Note that there are warranty implications of doing this -- you'd likely need to replace the drive in the case if you ever needed warranty service (not hard to do as long as you're CAREFUL and don't damage the case when removing the drive); and the warranty is 2 years vs. the 3 years that a bare drive comes with.   Nevertheless, it's not a bad option:  http://www.bestbuy.com/site/wd-easystore-8tb-external-usb-3-0-hard-drive-black/5792401.p?skuId=5792401

Ahhhh the same dilemma for me he here in Europe.  Harvesting a WD red 8 TB from external cost identical to the Seagate 8 TB archives drives.  Now I don't know anymore because money isn't the driving factor anymore in this comparison.

 

Edited June 9, 2017 by zonderling

[SSD]

RAID usage is very different that unRAID. RAID stripes data across all drives all the time. So every reasonably large write involves each disk in a somewhat chaotic random pattern under load. unRAID writes to each disk autonomously, and only parity writes flitter about. So unRAID (non-parity disk) writes are largely sequential.

 

I use the archive drives for functions that fill up disks over time, but have relatively slight deletes. I think it is a good use. For functions that require a fair amount of rewrite, I don't use them.

 

The archive drives ONLY use their cache on random I/O, not sequential. (Hence the reason they are a potentially poor choice for RAID.) So you can fill a disk with large files and never see the penalty. But I would not recommend it for storing a database where there is lots of random I/O. Although I believe that the PMR caches is something like 200G, so it would take a LOT of random writes to fill it up and be felt in terms of performance.

 

Overall, the Seagate PMRs perform very well. I personally would not think to use them for parity, but experience of others indicates that it is a valid use case and does not hamper performance. I will say that these drives are very fast for reading and writing, as compared to smaller drives. The higher density means more data per revolution. Makes a huge difference. I believe my 8T Seagates at ~5800 RPM are as fast or faster than my 7200 RPM 4T HGST drives.

 

One thought about using SMRs for parity or other high-write / overwrite scenarios is that the disk will be rewriting a lot more data, because it lays down multiple shingles per write. So a small write to the disk might result in a lot of read and write action to the disk. Not that I think the disk is going to wear out right away, but long term could affect longevity. Another one of these questions we'll never know the answer to.

[garycase]

45 minutes ago, bjp999 said:

I believe my 8T Seagates at ~5800 RPM are as fast or faster than my 7200 RPM 4T HGST drives.

 

That's actually easy to confirm.    The archive drives spin at 5900rpm (per Seagate), and have 1.33TB platters.

 

I assume your 4TB HGSTs have 1TB platters (most common).

 

So ... the HGST's take 8.33ms per revolution (at 7200 rpm)

      ... the archive drives take 10.17ms per revolution (at 5900 rpm) => which works out to 7.63ms for the same amount of data that the HGSTs will transfer in 8.33ms (since the platters are 33% denser)

 

So yes, the Seagates transfer data faster.

 

Of course if you're doing a lot of small transfers where the seek times are significant, then the HGSTs should be slightly faster, since they have slightly better seek times.

 

[SSD]

Thanks Gary!

 

Based on those numbers ...

 

The 7200 RPM 4Ts spin 22% faster rotating than the 8T 5900 RPM.

 

And yet in data transfer, the 8T drives process data 9% faster. 

 

Nothing to sneeze at for a "slow" drive!

[garycase]

... of course you could always buy 8TB WD Red Pro's or the HGST UltraStar He8's  -- 1.33TB platters AND 7200 rpm

 

[bonienl]

5 hours ago, zonderling said:

MTBF (hours) 800 000 Seagate Archive

 

LOL.. that is >90 years

[Squid]

3 minutes ago, bonienl said:

 

LOL.. that is >90 years

Especially since the MTBF of the average person is ~800 years.

 

http://www.qualityandproducts.com/2010/07/06/the-real-meaning-of-mtbf/

[SSD]

4 hours ago, garycase said:

... of course you could always buy 8TB WD Red Pro's or the HGST UltraStar He8's  -- 1.33TB platters AND 7200 rpm

 

 

HGST's 8T NAS are also 7200 RPM and I assume same density. They are a relative bargain at $269.

[garycase]

Simple fact is the 5900 rpm drives are PLENTY fast enough => anything that can sustain a data rate fast enough to saturate a Gb LAN is good enough for me.

 

... might change my mind when I (eventually) pick up a 10Gb LAN adapter