My new FreeNAS setup with some thoughts and unanswered questions

[Sqvirrel]

So after many years of looking at this forum and thinking about FreeNAS, I've finally bought a tower-server and got it running. My initial plan is to just use it as a network shared file server between the computers in my house. Later, maybe setup a datastorage for esxi, and maybe bareos storage daemon, if i get it running in a jail without too much hassle (anyone done it?)
Some questions I do have about my setup, and I have some thoughts about backup and storage that I wonder if anyone would comment on:

1) I use unRAID for archiving of my ripped movies and tv-series. I believe that this is sufficient redundancy, since its not critical data. The write/read speed is indeed a little slow, but does not matter for me here. Also, bit-rot would not be a problem of relevance here, would it? A few "rotten bits" here and there would only lead to some flashing bad frames anyway? Do people agree on this statement?

2) I've decided on going for full server-grade hardware on my primary FreeNAS system. Flashed IT-mode on the bundled LSI 1086e, running "Raid 0" on 2 disk right now, going to expand to "raid 10" as soon as I buy two more disks. Still I want more redundancy, and for the future, I will setup snapshots/replications. Next question: So lets say my primary FreeNAS server now works as it should with perfect intact data. Could my remote "replications-nodes" be cheap non-ECC FreeNAS servers? Yes, this is a cost-driven compromise... I have a lot of somewhat old consumer-grade hardware lying around. I could set up a few replication nodes at friends or family that receive intact data-replications from my primary FreeNAS. If the primary FreeNAS faults, I would have 2-3 remote nodes available. Do people say this could be acceptable?

3) I've followed the countless threads and posts about the never-ending argue about ECC/non-ECC. Now, lets say that some risky adventurer builds himself a non-ECC FreeNAS server. Would he be slightly more safe if he disabled the scrubbing? To prevent the (of course) faulty RAM to garbage up every bit and file while scrubbing through the disks?

 

[wblock]

A few "rotten bits" here and there would only lead to some flashing bad frames anyway?

It depends on where those bits are in the filesystem. If it's file data, maybe no problem. If the bad bits are in filesystem metadata, it could mean loss of access to those files, directories, or even entire volumes.

Flashed IT-mode on the bundled LSI 1086e, running "Raid 0" on 2 disk right now, going to expand to "raid 10" as soon as I buy two more disks.

In FreeNAS, "RAID0" is called a "stripe". The problem with the proposed layout is that if one disk in a stripe of single disks goes bad, the entire stripe is lost. If you want to use mirrors, start with them. So start with a mirror of your two drives. Later, additional mirrors can be striped with the first one, adding capacity. This is equivalent to starting with RAID1 (a mirror) and then going to RAID10 (a stripe of mirrors).

I could set up a few replication nodes at friends or family that receive intact data-replications from my primary FreeNAS.

Yes, this is a good approach. The more of these there are, the less I would worry about them requiring ECC RAM. If one loses some or all data, another will have those files. Right?

Would he be slightly more safe if he disabled the scrubbing?

Disable the protection from the very common bit rot to avoid the small possibility of RAM corruption damaging a volume? No.

ECC is just insurance. Non-ECC memory/motherboard/CPU might work fine for years. But it really hurts to lose valuable/irreplaceable data and realize that paying slightly more at the outset would have avoided it. Especially when you are planning on server-grade hardware which will support ECC. Another way to look at it is redundancy for RAM, to match the redundancy with hard drives.

 

[Sqvirrel]

It depends on where those bits are in the filesystem. If it's file data, maybe no problem. If the bad bits are in filesystem metadata, it could mean loss of access to those files, directories, or even entire volumes.

True true, well I can live with that risk since its just reproducible data (did I just jinx myself now?).

Yes, this is a good approach. The more of these there are, the less I would worry about them requiring ECC RAM. If one loses some or all data, another will have those files. Right?

Good good, then I will try to do something like this after my drives have arrived.

By the way, the first pool is going to consist of a total of 4 drives. I was mainly thinking of doing a stripe-mirror to maximize performance, BUT then it hit me, I'm running the server on a 1gbe connection. The transfer speeds will never get any higher than ~100 MB/s anyway? And a raidz2 will easily reach those speeds, wont it?

 

[Patrick Linley]

True true, well I can live with that risk since its just reproducible data (did I just jinx myself now?).


Good good, then I will try to do something like this after my drives have arrived.

By the way, the first pool is going to consist of a total of 4 drives. I was mainly thinking of doing a stripe-mirror to maximize performance, BUT then it hit me, I'm running the server on a 1gbe connection. The transfer speeds will never get any higher than ~100 MB/s anyway? And a raidz2 will easily reach those speeds, won't it?

RAID-5 for me, all the time, in all situations.


Sent from my iPhone using Tapatalk

 

[tvsjr]

RAID-5 for me, all the time, in all situations.

Then you must not be a FreeNAS/ZFS user... because ZFS doesn't support "RAID-5". Get used to saying "RAID-Z" - usually Z1, Z2, or Z3 (although Z1 is pretty much dead for larger drives).