![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
I currently run 3x 2TB Samsung F3EG in software RAID5 (mdadm on linux) & have been quite happy with it. However I'm down to ~200GB free so I started thinking about adding more disks, but then I remembered reading something a while ago about RAID5 being a bad idea for large arrays due to the rate of Unrecoverable Read Errors, so I started some research.
I had always assumed that if a disk failed, I could just replace it & let the array rebuild. I didn't realise that with the 1 URE in 10^14 that most consumer grade drives have, there would be a 35% chance of the rebuild failing, which kinda shocked me. On the positive side I discovered that the F3EG are actually rated 1 URE in 10^15 like enterprise grade drives, so there would actually only be a 3.5% chance of the rebuild failing. Better, but still a bit worrying.
Unfortunately it seems that the only disks you can buy now that are rated for 1 URE in 10^15 are proper enterprise ones which cost at least double their consumer counterparts. It seems that when Seagate bought Samsung's disk division they stopped making the F4EG with its 10^15 rating & replaced it with the Barracuda Green with 10^14. So if I buy new disks they will have to be 10^14 ones, as I can't really justify paying double.
So, what I'm thinking of doing is this;
1.) Build a RAID1 using 2x of my current 10^15 F3EG, on which I store all of the stuff that I *can't* just download again (eg my digital photographs, university work, etc.). I put the 3rd F3EG in an external enclosure & use it as a backup, which I sync once a week or something.
2.) Build a RAID6 out of 4x 2TB 10^14 disks, on which I store all of the stuff that I *can* just download again (eg TV shows, movies, anime, music, games). This I might not backup.
Does this seem like a sensible plan?
Now the other question I have, is what actually happens when a URE is encountered when trying to rebuild an array? Lets assume I stop using mdadm & instead buy a dedicated controller, then lets say that a disk in the RAID6 fails & there are then 2x UREs (the first RAID6 can recover from, the second it can't). Is the entire array lost, or is it possible to tell the controller to 'ignore' that bit & continue rebuilding the array but with a single corrupted file?
Edit - might I even be better looking at building a new box & using FreeNAS/BSD so I can use ZFS? Google tells me that ZFS will successfully rebuild an array in the face of multiple URE & provide a list of the corrupted files/block access to them.
I had always assumed that if a disk failed, I could just replace it & let the array rebuild. I didn't realise that with the 1 URE in 10^14 that most consumer grade drives have, there would be a 35% chance of the rebuild failing, which kinda shocked me. On the positive side I discovered that the F3EG are actually rated 1 URE in 10^15 like enterprise grade drives, so there would actually only be a 3.5% chance of the rebuild failing. Better, but still a bit worrying.
Unfortunately it seems that the only disks you can buy now that are rated for 1 URE in 10^15 are proper enterprise ones which cost at least double their consumer counterparts. It seems that when Seagate bought Samsung's disk division they stopped making the F4EG with its 10^15 rating & replaced it with the Barracuda Green with 10^14. So if I buy new disks they will have to be 10^14 ones, as I can't really justify paying double.
So, what I'm thinking of doing is this;
1.) Build a RAID1 using 2x of my current 10^15 F3EG, on which I store all of the stuff that I *can't* just download again (eg my digital photographs, university work, etc.). I put the 3rd F3EG in an external enclosure & use it as a backup, which I sync once a week or something.
2.) Build a RAID6 out of 4x 2TB 10^14 disks, on which I store all of the stuff that I *can* just download again (eg TV shows, movies, anime, music, games). This I might not backup.
Does this seem like a sensible plan?
Now the other question I have, is what actually happens when a URE is encountered when trying to rebuild an array? Lets assume I stop using mdadm & instead buy a dedicated controller, then lets say that a disk in the RAID6 fails & there are then 2x UREs (the first RAID6 can recover from, the second it can't). Is the entire array lost, or is it possible to tell the controller to 'ignore' that bit & continue rebuilding the array but with a single corrupted file?
Edit - might I even be better looking at building a new box & using FreeNAS/BSD so I can use ZFS? Google tells me that ZFS will successfully rebuild an array in the face of multiple URE & provide a list of the corrupted files/block access to them.
Last edited: