New Build - An exercise in overkill

[CoteRL]

I have a habit of going overboard when I build computers with the intent to not have to touch it again for years. This was the case when I ordered and installed all these new parts for my FreeNAS.

Case: CSE-836E26-R1200B
Mobo: X9SRL-F
CPU: Intel Xeon E5-1620 v2
RAM: Samsung DDR3-1866 16GB x 8 (128GB)
HBA: M1015 (Flashed IT Mode)
Disks: 6x 3TB WD Red Pro WD3001FFSX, 4x 3TB Seagate ST3000VN000
SLOG: Intel DC P3700 400GB (8GB partition)
NIC: Intel X540-T1 1x 10G, Intel E1G44E-T2 4x 1G

I have a single pool with a 10 disk RaidZ2 vdev, and after reading more and more I'm starting to think I managed to screw up when I chose to use RaidZ2 instead of mirrors. I am also not entirely sure I configured the SLOG correctly by only using an 8GB partition on the disk, but it seems to be working with no issues. I AM using NFS for an ESXi host with a few VMs so it is being used for sync writes (I've read enough to at least know that! :)).

The box serves mostly torrents, pictures and document backup, NFS datastore for a couple homelab VMs, and a backup destination for UrBackup which backs up my desktop and laptop.

I would love to hear if there are tweaks I can make to squeeze every drop of performance out of the system as possible. I got the parts I did with the intention/hope that the disks would always be the bottleneck of the system and as far as I can tell I've achieved that goal.

 

[Ericloewe]

RAM: Samsung DDR3-1866 16GB x 8 (128GB)

128GB? Holy mother of overkill.

 

[CoteRL]

128GB? Holy mother of overkill.

Yea.. Well, I kept reading more RAM, more RAM, more RAM.. so I got more RAM! I also wanted to give dedup a try on the backup and VM datasets just to see how it worked out so I was extra cautious with the 5GB for every 1TB of data to be deduped.

 

[jgreco]

128GB? Holy mother of overkill.

Really? 'Cuz I was thinking of putting 64-128GB of RAM on a system with only 4 8TB drives and going full dedup, possibly setting a record for RAM-to-number-of-disks ratio.

 

[danb35]

128GB? Holy mother of overkill.

I just bought that (more accurately, bought a server with that). But you're right, it's most likely overkill...

 

[CoteRL]

I'm glad to see I'm not SUPER crazy for doing it!

In regards to RaidZ2 vs Mirroring, is it even worth the trouble to move over to mirroring for this number of spindles/capacity? Is the performance difference that great to justify the work? I was thinking about it and I could get 6 more 3TB drives and create a new pool of mirrored vdevs, move the data over, destroy the original pool, and add those disks in as more mirrored vdevs. I would probably leave 1 or 2 disks as spare. I just don't know if any performance gain is worth the cost of the additional drives, the hassle of all the data moves, and the negligible increase in capacity (since I'm now losing 50% of my total instead of just 2 disks worth).

 

[jgreco]

For a truly small number of not-very-busy VM's, RAIDZ2 probably won't thrill, but it also won't kill.