I need some guidance on how I should build my nas configuration. I’ve been reading the forums and I have some ideas I just want to know people’s takes on what’s the best approach.
I’m converting an old dell workstation
Xeon 1620v3
16/32gb ram
Lsi 9207/9300 not sure which one - prices are close
16i or is 8i fine for this build.
I’ll be building this in a nas case that supports 12x 3.5 and 5x 2.5(ssd) might be able to fit a 6th
Run 2 raidz1 x 6(12 disks) in mirror or 12x raidz2. This is mainly for files for (vms, movies, personal files)
Run 2 raidz1 x 2(4drives) in mirror for vms this limits space but I like the redundancy it offers with performance or does it not matter on ssds?
Run 2 drives in mirror for boot.
Questions what’s the real performance hit of running in raidz2?
“Raidz1 in mirror” makes no sense.
VMs should be on mirrors, preferably SSDs, for efficiency with small blocks.
General, and large, files for NFS/SMB shares can be on raidz arrays, preferably raidz2 for safety and resiliency.
Xeon 1620v3
This should have sveral PCIe slots with bifurcation, so many opportunities for NVMe drives in place of or in addition to SATA SSDs.
Why? Is the NAS in a remote, hard to access, place?
9300-16i is basically 2x8i together with a switch. Double the power and electricity and will run hot.
Get 9300-8i, 9305-16i or 9207
1 Like
I would personally go for 12x RAIDZ2.
A mirror is not RAIDZ1 and you cannot create a 2x RAIDZ1 in the UI either. For zVols/virtual disks/iSCSI use mirrors, and if at all possible SSDs. If you use HDDs then you will need an SSD SLOG.
Also, only put the O/S on virtual disks. Access sequential data using NFS or SMB as normal files in a normal dataset in order to take advantage of asynchronous writes and sequential-prefetch.
Is this really worth it?
Finally…
Beware 2.5" HDDs as they are often SMR which should be avoided like the plague.
Here is a good “cheatsheet” for RAIDZ vs mirrors. And here is a pdf with a more in-depth overview.
Remember that IOPS is not the only performance measurement - for high-volume small e.g. 4KB random i/os with virtual disks/zVols/iSCSI or database it is important, however avoiding read and write amplification is even more important and that is why you need to use mirrors for these types of data.
But for sequential reads and writes then throughput is usually more important and RAIDZ2 is an excellent balance of redundancy overheads (much lower than mirrors) and sequential performance.
1 Like
