Building a cold storage server with Seagate Archive

[volume]

Hello!

 

I m planing to use UnRAID for a cold storage server for storing media.

 

Right now i have:

 

• 6 x Seagate archive 8TB
• an HP microserver (can take up to 6 drives)
• a desktop case from Fractal design R3 (can take up to 10 drives, maybe more with some customisation)
  with a cheap (~70€) GiGAbyte motherboard (it has 8 SATA and 2 eSATA) and a cheap (~40€) AMD CPU
  and 6GB of various rams from old computers that i had. i built it 4-6 years ago. so the hardware is pretty much old.

 

1) with 6 x seagate archive 8TB, due to the size of the pool (48TB) and the possibility of URE 

 

 

is the dual parity mandatory for pools with many TB (even if its build only on 6 drives) 

or its mandatory for large array of drives like 12 hard drives?

 

2) i will use the cold storage for ONLY storing media, nothing else...no VMs, no Plex etc

so it would be write once and only read occasionally (i also have a synology for 24/7 use)

 

do i need good and fast hardware?

or one of the cheapest cpu & motherboard with the most SATA and 4GB or 8GB ram would be good to go?

 

3) in case of rebuilding a pool on unraid, is the hardware responsible for rebuilding it quicker (thus safer for the other drives..)

or the bottleneck is the drive speeds?

 

 

Edited June 7, 2017 by volume

[zonderling]

I would invest in another non archive HDD for the parity function. For example WD Red 8 Tb. CPU and ram are sufficient for your purpose.


Sent from my iPhone using Tapatalk

[volume]

22 minutes ago, zonderling said:

I would invest in another non archive HDD for the parity function. For example WD Red 8 Tb. CPU and ram are sufficient for your purpose.


Sent from my iPhone using Tapatalk

 

Could you please explain the benefits that i will gain if i have the WD RED 8TB for parity?

it will give me faster rebuilt times?

 

Should i go with dual for an array of 48TB+ (6 x Seagate 8TB) ?

Edited June 7, 2017 by volume

[trurl]

2 hours ago, volume said:

is the dual parity mandatory for pools with many TB (even if its build only on 6 drives) 

or its mandatory for large array of drives like 12 hard drives?

 

Single parity allows data to be calculated and reconstructed if there is a single "missing" drive.

Dual parity allows data to be calculated and reconstructed if there are 2 "missing" drives.

 

So, the need for dual parity depends on the number of drives, not the total capacity. The more drives the more risk for 2 "missing". Dual parity (and indeed single parity) are not "mandatory", it just depends on how much risk you want to allow.

 

A drive can be "missing" from the array even if it is still physically present and working. unRAID disables a disk (and will not use it until rebuilt) whenever it has a write failure. The failed write (and any subsequent writes) still update parity so the contents of the physical disk are invalid (until rebuilt) but the valid contents (the emulated disk) are still in the array.

 

The point made in the video is that during a rebuild of a single disk with single parity, if another disk has a problem the rebuild is compromised and may even fail to complete. Dual parity allows the rebuild to succeed. Of course, if more than 2 drives have problems simultaneously, dual parity will not solve this, and unRAID doesn't allow more than 2 parity drives. In any case, since each disk is an independent filesystem, files on the other disks are not affected.

[volume]

39 minutes ago, trurl said:

 

Single parity allows data to be calculated and reconstructed if there is a single "missing" drive.

Dual parity allows data to be calculated and reconstructed if there are 2 "missing" drives.

 

So, the need for dual parity depends on the number of drives, not the total capacity. The more drives the more risk for 2 "missing". Dual parity (and indeed single parity) are not "mandatory", it just depends on how much risk you want to allow.

 

A drive can be "missing" from the array even if it is still physically present and working. unRAID disables a disk (and will not use it until rebuilt) whenever it has a write failure. The failed write (and any subsequent writes) still update parity so the contents of the physical disk are invalid (until rebuilt) but the valid contents (the emulated disk) are still in the array.

 

The point made in the video is that during a rebuild of a single disk with single parity, if another disk has a problem the rebuild is compromised and may even fail to complete. Dual parity allows the rebuild to succeed. Of course, if more than 2 drives have problems simultaneously, dual parity will not solve this, and unRAID doesn't allow more than 2 parity drives. In any case, since each disk is an independent filesystem, files on the other disks are not affected.

 

thanks!

 

I'm afraid that if i start with single parity (5 x 8TB disk for media, 1 x drive parity) and add another drive for parity in the future, the risk of failing to rebuilt the whole array (40TB), will be high due to the URE error. is that true or i missing something?

 

do we have an estimate on how many hours/days will take to rebuilt such a big array?

 

its a good idea to have "better/faster" drives for parity like the WD RED?

Edited June 7, 2017 by volume

[trurl]

28 minutes ago, volume said:

do we have an estimate on how many hours/days will take to rebuilt such a big array?

Assuming not controller bottlenecks, the time to rebuild is mostly a function of the size of the single disk, since all disks are read in parallel for rebuilds. I haven't used 8TB disks and haven't needed to do a rebuild recently, but I can get a 6TB parity check done in about 15 hours. Rebuilds would be similar, maybe even a little faster.

[JonathanM]

45 minutes ago, volume said:

 

I'm afraid that if i start with single parity (5 x 8TB disk for media, 1 x drive parity) and add another drive for parity in the future, the risk of failing to rebuilt the whole array (40TB), will be high due to the URE error. is that true or i missing something?

Mostly FUD. The chance is certainly there, to be sure, but if it were a common occurrence, the forum would be littered with horror stories and we would all be telling you to never trust a large array without dual parity.

 

Now... that said, URE's are hardly the biggest threat to your data. Much more common events will cause data loss, so a MUCH better strategy is to use the drive capacity that would be devoted to a second parity disk and use it to keep offline backups of your critical files.

 

If the only copy of the file is on your unraid server, you are at risk. Single parity, dual parity, doesn't matter.

[volume]

17 hours ago, jonathanm said:

Mostly FUD. The chance is certainly there, to be sure, but if it were a common occurrence, the forum would be littered with horror stories and we would all be telling you to never trust a large array without dual parity.

 

Now... that said, URE's are hardly the biggest threat to your data. Much more common events will cause data loss, so a MUCH better strategy is to use the drive capacity that would be devoted to a second parity disk and use it to keep offline backups of your critical files.

 

If the only copy of the file is on your unraid server, you are at risk. Single parity, dual parity, doesn't matter.

I understand, but unfortunately i don't have the ability to back up all my media due to many TB.

So i try to find the hot spot of keeping my data safe, without breaking the bank.

 

Unfortunately i can't afford to buy 6 x WD RED 8TB hard drives, but if its a good idea to have one WD RED 8TB for parity instead of the Seagate archive 8TB, please let me know.

or others tips for making my cold storage server "safer"

 

I also read that i need a good quality power supply, any suggestion for model?

thanks 

[SSD]

18 hours ago, jonathanm said:

If the only copy of the file is on your unraid server, you are at risk. Single parity, dual parity, doesn't matter.

 

I agree with this. With a large media server, much of the data is available through alternate sources. You could re-rIp, re-download, re-purchase that our similar content. PITA? Absolutely. But possible for media you really care about. But truly unique works like photographs, home movies, and personal files have no alternate recovery method, and likely only consume a small portion of your array. At a minimum these files should be backed up. 

 

The number one risk to unRaid data is not drive failures but cable failures. It is very common that, while replacing or adding a disk, that another cable is nudged, creating a marginal connection, and resulting in that disk going offline intermittently. What often happens is a user will replace a disk (to upside or replace a disk with smart issues), and during the rebuild, another drive drops due to the cable being loose. This can cause a dual drive failure scenario, when in fact both the disk being replaced and the disk that dropped offline are both fully readable. Dual parity might smooth this recovery process, but no data loss should occur either way.

 

But the better way to protect from this scenario is to use drive cages and locking cables where possible. This allows drives to be exchanged or added without sticking your hand inside the case. And virtually eliminates this most common problem.

 

I did a back of the napkin computation on dual parity, and the protection it provides is virtually the same as the chances of house fire. A small fraction of 1% per year disk on a big array. And my assumption going in one was that an array would have a real drive failure every year. Not trying to brag or tempt fate, but I have never had a drive fail that required a parity rebuild to recover the data. And I've been using unRaid for 10 years. (But I do replace disks that develop smart issues, and preclear every disk before adding them to the array. Both good idea to avoid drive failures.) All my worst problems have been self inflicted mistakes or cabling problems caused by side effects of replacing disks. And the best reason to have dual parity is actually to protect yourself from yourself.

[JorgeB]

I use a 8TB archive disk as parity on one of my servers and it's been working great, rebuilds, parity checks, etc will be the same as using a RED for parity, a RED may perform better as parity if writing simultaneously to more than one disk, but that's not important for me.

[trurl]

1 hour ago, volume said:

i don't have the ability to back up all my media due to many TB

The only reason I have backups of my "re-obtainable" files is because after various upgrades to my main server over the years, including upsizing disks, I had enough hardware leftover to build a backup server. At first, I didn't have these backups.

 

I have always had multiple backups of my "un-obtainable" files though. The original files are on my PC, and unRAID is the backup. And I have rotating external drives for offsite backups of these files.

 

You don't have to backup everything, but you must have a backup plan.

[volume]

yes the "un-obtainable" files are on my PC, in my synology nas, and it will be on unRAID when i finally built it.

But most of the "re-obtainable" files are quite hard to find (with custom subtitles etc) and time consuming to re-rip my collection. 

obviously i don't mean occult video, snuff movies or porn..

 

what cables do you recommend or case?

does the fractal design define R3 good enough?

 

thanks!

[trurl]

42 minutes ago, volume said:

most of the "re-obtainable" files are quite hard to find

So you can include these in your backup plan. I backup some of mine on my offsite also. Some of my music would indeed be hard to re-obtain. My externals offsite have plenty of capacity for these. It was mostly video that I didn't backup until I built my backup server.

[SSD]

1 hour ago, volume said:

what cables do you recommend or case?

does the fractal design define R3 good enough?

 

There is no brand name cable that is considered the best that I am aware of. I buy my cables from Monoprice. They have very good prices and I have been happy with the quality. I bought a bunch of locking cables from them several years ago and still have a good supply. They work nicely. They are "IO Gear" branded, although when I bought them it did not advertise a brand name.

 

As far as the case, There are 5 basic options ...

 

1 - A big rack mounted monster. Norco or Supermicro (often obtained on ebay). These have the drive cages built in, and are well-priced and relatively popular for big arrays. I personally do not have a rack to mount such a thing. So for me it would take up a lot of floor space. I hear a lot of complaints about the noise levels, but is a good option for many.

 

2 - A case that have a bunch of 5.25" drive slots, for example Antec 900/1200. You can install 3/4 5in3 cages (like SuperMicro CSE-M35T-1B) for 15/20 drives. Not as easy to find any more. I have a Sharkoon Rebel 12 case like this, and I love it. Small footprint, easily moved, good cooling. And an Antec 900 for my backup server. Both work great. I like this style case with cages best of all.

 

3 - A tower case with some accommodation for drives. They angle them for easier access to the cable connections, some sorts of rails or trays, and may include some cable management features. I'd put the R3 in this category. They are not my favorite because you are still having to take the computer apart and touch the cables every time you want to swap drives. Risk to the cabling is reduced but not eliminated. A big change involving 3 or 4 drives I can do in 5 minutes with 0 risk, would easily take an 30 mins to an hour and could easily result in a dropped disk and need to rebuild. This type of case may be ok for smaller numbers of disks - maybe 4-5, but I am not a big fan of loading up such a case with lots of drives.

 

4 - A traditional case with no cages or special accommodates for drives. You screw them into the case, maybe using adapters to mount 3.5" drives in 5.25" slots. These are awful! Accessing cabling is difficult. You are going to constantly have cable problems. Every time you add or exchange disks you will create a risk of creating intermittent problems which result wasting nights and weekends to sort it out. Using it for a 3 drive array is ok, but I'd not try to do more. 

 

5 - An innovative approach would be to buy the 5in3s and keep them external to the case. My backup server with 15 drives isn't big enough, so I have 2 5in3s sitting nearby powered by a PSU pigtail, and utilizing a controller with external SFF-8088 plugs. I'm thinking you could take four of them and create a "quad" that gets screwed together with some ingenuity and use that for ALL of your array disks. The computer itself could be small and perhaps only include the SSD, USB disk, and controller card(s).Not every person could accommodate such a configuration, but I think it is viable in many situations. And it would be almost portable!

 

A note on the 5in3 cages. Drives are 100% externally replaceable. You don't even see a wire. Replacing a disk means shutting down, removing the disk tray, unscrewing 4 screws that secure the drive to the tray, installing the new disk into the tray with the 4 screws, and sliding it into the cage. Takes 2 minutes. You know you have not rattled any wires. There are other advantages. While one drive is not that heavy, put 15 or 20 of them in a case and it can give you a hernia. With the drive cages, , you can literally remove every drive, and then the server is not too heavy to pick up and move to a workbench or relocate to another location. You can then carry the drives separately and reinsert. 

 

Most of these units support hotswap, meaning you can add or remove to a running server. But I never use them for hotswap, I shutdown and do what I need to do, without opening the case. 

 

Cheers!

[volume]

2 hours ago, bjp999 said:

 

There is no brand name cable that is considered the best that I am aware of. I buy my cables from Monoprice. They have very good prices and I have been happy with the quality. I bought a bunch of locking cables from them several years ago and still have a good supply. They work nicely. They are "IO Gear" branded, although when I bought them it did not advertise a brand name.

 

As far as the case, There are 5 basic options ...

 

1 - A big rack mounted monster. Norco or Supermicro (often obtained on ebay). These have the drive cages built in, and are well-priced and relatively popular for big arrays. I personally do not have a rack to mount such a thing. So for me it would take up a lot of floor space. I hear a lot of complaints about the noise levels, but is a good option for many.

 

2 - A case that have a bunch of 5.25" drive slots, for example Antec 900/1200. You can install 3/4 5in3 cages (like SuperMicro CSE-M35T-1B) for 15/20 drives. Not as easy to find any more. I have a Sharkoon Rebel 12 case like this, and I love it. Small footprint, easily moved, good cooling. And an Antec 900 for my backup server. Both work great. I like this style case with cages best of all.

 

3 - A tower case with some accommodation for drives. They angle them for easier access to the cable connections, some sorts of rails or trays, and may include some cable management features. I'd put the R3 in this category. They are not my favorite because you are still having to take the computer apart and touch the cables every time you want to swap drives. Risk to the cabling is reduced but not eliminated. A big change involving 3 or 4 drives I can do in 5 minutes with 0 risk, would easily take an 30 mins to an hour and could easily result in a dropped disk and need to rebuild. This type of case may be ok for smaller numbers of disks - maybe 4-5, but I am not a big fan of loading up such a case with lots of drives.

 

4 - A traditional case with no cages or special accommodates for drives. You screw them into the case, maybe using adapters to mount 3.5" drives in 5.25" slots. These are awful! Accessing cabling is difficult. You are going to constantly have cable problems. Every time you add or exchange disks you will create a risk of creating intermittent problems which result wasting nights and weekends to sort it out. Using it for a 3 drive array is ok, but I'd not try to do more. 

 

5 - An innovative approach would be to buy the 5in3s and keep them external to the case. My backup server with 15 drives isn't big enough, so I have 2 5in3s sitting nearby powered by a PSU pigtail, and utilizing a controller with external SFF-8088 plugs. I'm thinking you could take four of them and create a "quad" that gets screwed together with some ingenuity and use that for ALL of your array disks. The computer itself could be small and perhaps only include the SSD, USB disk, and controller card(s).Not every person could accommodate such a configuration, but I think it is viable in many situations. And it would be almost portable!

 

A note on the 5in3 cages. Drives are 100% externally replaceable. You don't even see a wire. Replacing a disk means shutting down, removing the disk tray, unscrewing 4 screws that secure the drive to the tray, installing the new disk into the tray with the 4 screws, and sliding it into the cage. Takes 2 minutes. You know you have not rattled any wires. There are other advantages. While one drive is not that heavy, put 15 or 20 of them in a case and it can give you a hernia. With the drive cages, , you can literally remove every drive, and then the server is not too heavy to pick up and move to a workbench or relocate to another location. You can then carry the drives separately and reinsert. 

 

Most of these units support hotswap, meaning you can add or remove to a running server. But I never use them for hotswap, I shutdown and do what I need to do, without opening the case. 

 

Cheers!

 

Since that i m stating with 6 disks i will go with the 3 option and stick with my Define R3 at the moment.

And when i will add more drives, i will consider to move to option 2 with the 5in3 cages. 

 

thanks!

[SSD]

Fair enough. Good luck with your build!

[zonderling]

On 6/7/2017 at 3:04 PM, volume said:

 

Could you please explain the benefits that i will gain if i have the WD RED 8TB for parity?

it will give me faster rebuilt times?

 

Should i go with dual for an array of 48TB+ (6 x Seagate 8TB) ?

WD RED or even GOLD or any other disk that is not based on the shingled magnet recording technology.

 

Your parity drive is the only drive in your array that will sustain heave write operations. Each bit that is written in the array makes a write on this parity disk.

This type of drive leverages a lot on its internal cache ( about 20 GB ) to do houskeeping on incomming writes.  A sort of temporarily parking spot for data that needs to be written to disk.  By design in parks the data in cache, reads the data on another spot on the disk, then writes boht data as a sort of combined striped data.  That's how Seagate gets so mucht TB on the platter.

 

By design your parity disk will always be 100% full so performance will degrade.  And since the array data disk speeds is determined by the write speed of your parity disk, it would make sense to have a normal 8 TB for parity and leave your archive disk for what they are intended for.  Few writes, many reads.

 

[zonderling]

On 6/8/2017 at 0:00 PM, volume said:

Unfortunately i can't afford to buy 6 x WD RED 8TB hard drives, but if its a good idea to have one WD RED 8TB for parity instead of the Seagate archive 8TB, please let me know.

It makes sense.  Motivation in my previous post.  Basically Seagate Archives are not good for writing a lot to.  They will get slower, slowing down your array, and they will fail fast.

[BRiT]

It doesnt matter, the seagate archive drives function perfectly fine for parity drives even for main array operations where you wont be randomly modifying the data. The drives have been in operations on numerous arrays ever since they were available and not one person has ever hit the write penalty SMR wall.

[volume]

Yes i will not modify the data.
I will full all the drives and only occasionally, i will copy the files that i need, to my main server.

[JorgeB]

32 minutes ago, zonderling said:

and they will fail fast.

 

This is based on what? alternative facts?

 

From other users on the forum an my own experience this is not true, I have 5 of theses disks (one of them as parity) for over a year with 0 issues, and see no evidence that these disks have an higher fail rate than any other disk.

[zonderling]

7 minutes ago, BRiT said:

It doesnt matter, the seagate archive drives function perfectly fine for parity drives even for main array operations where you wont be randomly modifying the data. The drives have been in operations on numerous arrays ever since they were available and not one person has ever hit the write penalty SMR wall.

Ive got and use 8TB Seagate disk as well and I use them for what they are intended. Archiving .... Cold storage. For me this is to be taken literary.  I take backup by external docking bay, then remove it for offline, cold storage.

 

I do notice a big performance hit when writing a lot of sequential data ( ie backup).  When its new, clean and formatted I get write speed in excess of 100MB/s when its 3/4 full it drops down below 30MB/S

 

Ive had it a few days in my "production" rig as parity at i remember slow performances as well ( again 30 MB/s all the time.

 

Articles about this drive shingling seems to confirm my theory.  

 

I'm not saying it will not work, I'm sure a lof of use cases will be here where it works, I'm only answering to the topic starter request to optimize his setup.

 

[zonderling]

8 minutes ago, johnnie.black said:

 

This is based on what? alternative facts?

 

From other users on the forum an my own experience this is not true, I have 5 of theses disks (one of them as parity) for over a year with 0 issues, and see no evidence that these disks have an higher fail rate than any other disk.

 

http://www.storagereview.com/seagate_archive_hdd_review_8tb

 

 

RAID Usage with SMR

With the attractively low price per TB that the Seagate Archive 8TB HDD has, it can be difficult to not consider purchasing a set for NAS storage. StorageReview strongly recommends against such usage, as at this time SMR drives are not designed to cope with sustained write behavior. Many contend that NAS shares tend to be very read-focused during normal operation. While that's true, the exception is when a drive fails and a RAID rebuild has to occur. In this case the results clearly show that this implementation of SMR is not a good fit for RAID. 

[JorgeB]

3 minutes ago, zonderling said:

RAID Usage with SMR

 

unRAID is not RAID, and even if you use with RAID it may not be the best performer but there's no evidence it will fail faster.

[zonderling]

19 minutes ago, johnnie.black said:

 

unRAID is not RAID, and even if you use with RAID it may not be the best performer but there's no evidence it will fail faster.

I new you where going to say that

 

I know its not RAID (hence the name ) but it is similar in the sence that parity disk is aways written to, similar to RAID arrays

 

But if you need cold figures there is a 1/5 difference in expected life cycle.

 

MTBF (hours) 1000 000 WD RED
MTBF (hours) 800 000 Seagate Archive

 

Also you would agree for every write action on the platter, the disk needs to do 3 actions.  Normal wear and tear would suggest it will have an effect on expected MTBF

Edited June 9, 2017 by zonderling