Freenas learning build

[Nimbu]

Hi guys,

Long time lurker first time posting, so please be gentle!

My synology DS developed the atom bug and is about to be RMA'd. I've pulled off my data so I am good on that front, but the experience has me thinking to explore other options.

Over the years I have had all sorts of solutions. From a HP mediasmart ex495 running original WHS2003, then whs2011, server 2012 with stablebit drivepool, to NL54 running xpenology to finally a DS1815+.

I have always wanted to learn and mess around with ZFS but the old GUI put me off as well as an old boss who insisted I used OpenIndiana and do everything via the CMD line! I'm not stranger to it, but over the years I've got lazy and like shiny GUI's.

Anyways, before I go all out and attempt to build a power house device I thought it would be a good idea to setup a testing box so I can skill up a bit. The final intended use is a home server that will serve my media to my endpoints. Plain old samba shares, no plex etc.

I am planning to use the following for my test rig.

HP Microserver Gen 8
8GB ECC UDIMM
4 x 6TB (Config undecided just yet)

Given it's a testing rig, should I really consider 16GB ram? Or will the 8GB do me fine for now?
Anything else hardware wise I should consider?

For the disk setup, I'm undecided between RAIDZ1 or RAIDZ2 or the 2+2 mirror / stripe.

Home network is only 1GB and generally only one endpoint will be accessing the media files at a time so performance wise I think I'd be ok.

However redundancy wise I'm concerned, nothing will ever be on the box that I don't have backed up or is non replaceable, but it would take me a while to repopulate my media from original disks so I'd like to avoid that.

The idea of a URE during a resliver scares the crap out of me, so I am thinking Z2 / 2+2. From my limited understanding the 2+2 would be much faster so the time the pool will be degraded is lower, is that correct?

Is the following true that in a bigger pool, say 8 disks, if one fails and I reslivering that:

1) it would be faster as I am rebuilding from 7 disks ?
2) because I'm reading from 7 disks I would theoretically be less likely to hit a URE?

Since later I will be having a bigger pool, is the process by which replacement disks are processed the same? I.e. I plan to simulate drive fails by pulling and rebuilding? In my final NAS I will be using 8 x 6tb so would more than likely go Z2.

Please don't forget this is just a testing rig so I can learn freenas before I build my monster NAS, which I already have the following hardware for:

Supermicro S2011 mobo (forget the model now)
E5 1320 (I think)
96gb ram ddr3 ecc registered
8 x 6tb disks
LSI based HBA flashed to IT mode.

 

[MatthewSteinhoff]

Given it's a testing rig, should I really consider 16GB ram? Or will the 8GB do me fine for now?

Eight is fine for testing. With just a few users, it might even be fine to use. RAM is easy to add after the fact. Deploy cheap and upgrade as needed based on actual performance.

For the disk setup, I'm undecided between RAIDZ1 or RAIDZ2 or the 2+2 mirror / stripe.
2+2 would be much faster so the time the pool will be degraded is lower, is that correct?

RAIDZ1 is depreciated except for tiny data sets which you have already exceeded. With only four drives, do a striped mirror. It will give you the exact same amount of space and twice the performance. So, yes, with four drives, going striped mirror is the right choice.

Is the following true that in a bigger pool, say 8 disks, if one fails and I reslivering that:
1) it would be faster as I am rebuilding from 7 disks ?
2) because I'm reading from 7 disks I would theoretically be less likely to hit a URE?

1. I don't think so. RAIDZ is as fast as the slowest drive and, since every drive must be read from and written to, I'm not sure how more drives would make the rebuild any faster.

2. The more drives, the larger the drives, the greater bits in play, the greater chances for error. It's math.

If you are deploying more than four drives and care more about storage efficiency than performance, go with RAIDZ2. That will greatly reduce the chances your pool will fail on rebuild. Of course, no level of RAID is a substitute for a good backup.

Cheers,
Matt

 

[Ericloewe]

Resilvers used to be seriously IOPS-bound. With sequential resilver, that is less the case now.
Wider vdevs help speed things up if you can read huge contiguous chunks at once, but I won't hazard a guess as to whether that offsets the increase in data for the same IOPS figure (each vdev has about the IOPS of a single disk).