RAID options

[Rampersandall]

Planning a new build, posting in this thread because my question is about RAID.  I've never done a raid setup before - just watched Linus and NCIX vids introducing me to the different arrays.  Most interesting to me is RAID 5.  Still unsure about something.

 

Ultimate goal for the new build is use as a plex server.  Ubuntu if possible, but second choice is Windows 10.  It's going to be an "always on" system - I'm searching around now trying to find pro's and con's to each setup still.

 

I own two drives now.  But definitely intend to expand over the years.  I can reasonably see 4-5.  But anyway.  

 

I would like to know if there's a raid setup with "drop-in" expansion.  IE, add a 8TB drive and your 40tb becomes 48.  give or take a few TB.  Is that possible with R5?

 

Also, are there any benefits to having an internal NAS, over one of those external, standalone NAS boxes?

 

Regards,

Ran

[trag1c]

Too my knowledge expansion of a raid regardless of level is based upon what the controller supports. Some may allow you to enlarge the array by adding more drives of the same capacities and some may not. This could also be that the controller may allow you add more drives to a raid 5 but not a 6 as an example. 

However I would strongly recommend against RAID 5 as it is vulnerable to failure because you only have a failure buffer of one drive. So if a second drive fails you lose everything. If you have to rebuild an array after losing a drive then you're in a position where another drive failure has a high probability of failing before you can actually rebuild the array causing you to lose everything.

 

RAID 6 has slightly more mitigation against this but still suffers from the same problem. (technically they all do) I would actually recommend RAID 10 (1+0) in which you have a mirroring of drives and a stripping of data across them so you in essence get super fast reads and super fast writes. Another benefit is because the disks are mirrored when have a failure you only have to rebuild the mirror which is a simple copy operation compared to RAID 5 or 6 in which you have to spin up all the drives to read data from which to make calculations to write data on the new drive. So with RAID 10 your far less likely to have another disk failure while you're rebuilding the array because you're not hitting the disks as hard or for as long. As far as the amount of disks that can fail it varies. If you have a total of 4 disks of which you have 2 mirrors with 2 disks each in a raid zero then you can lose 1-2 disks. 1 per mirror. You can scale the failure mitigation by how many disks are in the mirrors and how many mirrors you have.

[Kilrah]

RAID is rigid, once set up no changes (apart from adding a completely new, independent array of course).

You might want to look at Unraid or Drobo NASes, those support plug and play drive replacement and expansion with anything at anytime.

If you really care about your data you will want a 2nd completely independent backup solution in addition to whatever redundancy is built into your main storage so having more than one drive failure tolerance isn't really useful. 

[Rampersandall]

Thank you - I appreciate the feedback :)

 

I am definitely into the idea of faster reads and writes.  So I'll plan for R10.  If I understand this correctly... for R10, If I wanted say 20tb of media storage - I'd need 40tb of physical?  

 

And the only reason I'm leery of dedicated server OS's like unraid, is because I'm very unfamiliar with them.  I'll do some investigating.

 

While I do care about my data... there's nothing on my plex drive I couldn't easily just replace.  I don't want to... but I could.  All my important docs, saved files, etc... those are backed up to a google drive, a thumb drive, an external, and an unused closet PC that's packed into a box.  If one day Phenom II based system are valuable, I have a gold mine in there.  Movies, TV Series and Music.  That's all that's going on the server drives.  Games will be stored on separate SSD's.  Which I'm not certain about yet, but I may also RAID together.  It's a future build.  All I have now is a MOBO my brother in law gave me which I'm trying to make use of.  It supports 6th gen intel CPU's.  So I'll probably go with the 6700k.

[Electronics Wizardy]

2 hours ago, Rampersandall said:

d the only reason I'm leery of dedicated server OS's like unraid, is because I'm very unfamiliar with them.  I'll do some investigating.

unraid is very easy to use, and. gret soluion here. You can just add drives and get more capacity.

 

2 hours ago, Rampersandall said:

I am definitely into the idea of faster reads and writes.  So I'll plan for R10.  If I understand this correctly... for R10, If I wanted say 20tb of media storage - I'd need 40tb of physical?  

The performance difference won't matter here, you will be network limited anyways.

 

 

[Nick7]

3 hours ago, trag1c said:

RAID 6 has slightly more mitigation against this but still suffers from the same problem. (technically they all do) I would actually recommend RAID 10 (1+0) in which you have a mirroring of drives and a stripping of data across them so you in essence get super fast reads and super fast writes.

I have to disagree.

RAID6 is the most safe 'standard' RAID type. More safe than RAID10 - where if you have bad luck double disk failure means game over.

While RAID6 rebuild time is longer, it's still more safe than RAID10.

Performance-wise, reads are excellent on RAID6, but writes suffer.

Nonetheless, on any new storage system - RAID6 is considered nowdays default option, for a reason.

There's also the capacity question, where RAID6 again wins.

RAID10 is basically only used if you are using HDD's and require decent write speed.

 

For home-use, and storage-type NAS, look no further than RAID6.

[trag1c]

2 hours ago, Nick7 said:

I have to disagree.

RAID6 is the most safe 'standard' RAID type. More safe than RAID10 - where if you have bad luck double disk failure means game over.

While RAID6 rebuild time is longer, it's still more safe than RAID10.

Performance-wise, reads are excellent on RAID6, but writes suffer.

Nonetheless, on any new storage system - RAID6 is considered nowdays default option, for a reason.

There's also the capacity question, where RAID6 again wins.

RAID10 is basically only used if you are using HDD's and require decent write speed.

 

For home-use, and storage-type NAS, look no further than RAID6.

The chances of suffering failure of multiple disks in the exact some moment is basically non existent but the chance of another disk failing shortly after another one is significantly more probable. Where you run into irrecoverable failures is almost always on the rebuild of an array. This is where RAID 10 really shines above 5 or 6 for the simple fact that all that happens during the rebuild of RAID 10 is a data transfer from one mirror drive to the other mirror drive. RAID 5/6 levels rebuild times are complicated by array and drive size. In RAID 10 the time it takes to rebuild is simply the average sustained write speed to the replacement drive from the other drive(s) in the mirror but in RAID 5 or 6 you have to calculate what the data is going to be on the replacement drive. So instead of simple transfer of data you are left with placing heavy sustained load on every single drive in the array to calculate the new data that has to be written to the replacement drive that will take days to finish. So the biggest window of vulnerability to the array is there for days from which at any point more drives can drop from the array and cause a complete array failure or even longer period of days for which more drives can fail leading you back to total failure. The other issue is if you encounter an unrecoverable read error or in other words a bad sector. One bad sector during a rebuild could destroy an entire array if it was RAID 5. RAID 6 can suffer a disk failure and an URE and still rebuild but the problem is the correlation between one disk dying and another. Now you throw in two disk failures and then you get an URE you're done and all of your data is gone. URE's typically occur once in every 200,000,000 sectors (or around 12TB) so now you have an op that is mentioning 40TB so you will hit at least 3 URE's while you rebuild your array and if you have an additional drive failure during the rebuild which is not out of the realm of possibilities you will lose everything.

 

RAID 10 can have more than one drive drop from the array even in a simple 4 drive configuration. The only time that you can't have more than one drive fail is if its in the same mirror from which there were only 2 drives to begin with. But as I stated you're only vulnerable for hours compared to days and you're still left with much higher probability of success because you can have drives drop out in other mirrors but as long as its not the one that's rebuilding you still have your data. Where as in 5 or 6 any drives dropping can be the contributing factor for an irrecoverable data loss. When it comes to URE's and RAID 10 its a no big deal because at worst you lose that sector and keep copying.  You can restore the bad data from back up later.

 

As far as RAID 10 only being for HDD's that's total crap as you get a benefit as long as you have the network bandwidth to support it. Especially small files can work be

RAID 6 is going to the way of the dodo (and very quickly doing so just like RAID 5) because of drive capacities. I highly doubt when they first patented RAID 5 back in 1986 that they had any clue that you would have up to 18 or 20TB drives available to consumers let alone the bigger drives that will be accessible to enterprise. RAID was never a solution for back up but as far as trying to keep any sort of sense of up time you would be going RAID 10 or some more nested derivative of it all because of relationship between drive capacities, rebuild time, and unrecoverable read errors.
 

[Nick7]

11 hours ago, trag1c said:

*snip lots of stuff*

I have to disagree completely.

 

First, URE's you mention are stated in some article almost several decades ago. By what you claim - on average each 12TB you should get an URE. That's Unrecoverable Read Error. If you have storage system which does on average 1GByte/sec reads/writes, that would mean you get URE's each 3.5 hours, which is simply not true.

On storage systems each URE is logged, and you are ware they happen. I work with storage systems of size of 1+PB, and those kind of errors are really scarce, and generally would mean disk is dying. Claiming 1 URE each ~12TB is quite literally a FUD.

 

As for double disk error in RAID10 -  I actually did have this issue once some 15 years ago. Result - some data was lost, but there's backups for that kind of situation. Not fun thing anyway.

 

As for URE's and  chanses of disk failure, something a bit more newer from IBM:

https://www.ibm.com/support/pages/re-evaluating-raid-5-and-raid-6-slower-larger-drives

 

I will not go much into discussion about what is says there, but I suggest you read it. Will spark some more realistic data about probability of double disk errors even on slower and larger NL (aka SATA) drives, and RAID5/RAID6 probability of double/triple disk failure and/or URE's during rebuild.

 

Lastly, on storage systems as IBM, EMC, Fujitsu and similar, default RAID is - RAID6. For Flash drives they still decide that RAID5 is good for that. No RAID10.

 

With large drives, RAID10 is only used when you require decent write speeds, otherwise it's always RAID6.

 

Spoiler (includes URE's and drive failure rates):

Drive type	RAID-5 rebuild failure (percent)	RAID-6 rebuild failure (percent)
600GB 15K rpm	0.397	0.0000861
6 TB 7200rpm	4.12	0.0163