hey everyone… I’m supposed to be getting some money that i’m okay allocating to upgrading my storage setup. I currently have 2 Dell R720XD’s, one running TrueNas Core, the other Scale. They are on small drives, (6TB and 8TB), but I’m getting a new server and new drives.
Here’s the new (to me) hardware:
- Server: 4U 36 Bay SAS3 12Gb/s TruNas Storage Server Xeon 24 Cores 256GB 2x 2.5" SQ PS | eBay (yes it’s OOS for now, but they keep adding quantity as they configure more for shipping)
- Main HDDs: Seagate Exos X16 16TB SAS 12Gb/s 3.5" Enterprise HDD (ST16000NM002G) | eBay
- U.2 adapters: https://www.amazon.com/gp/product/B09WR1ZZVQ
- U.2 drives: Intel P4510 2TB PCIe Gen3 x4 NVMe U.2 2.5" Enterprise SSD - SSDPE2KX020T801 | eBay
- Network adapter: Intel 520-DA2 10gig dual SFP+ that I have on hand
My plan is to have 4 data VDEV’s of 6 16TB HDDs each, RAIDz2. I’m also going to have 2 hot spare HDDs and a few additional cold storage HDDs.
I’m a little confused on optimizing SLOG for synchronous writes, and L2ARC… my thought is to use mirrored U.2’s for each, and have a couple of cold storage extras. But I also read an article discussing the poor performance of SSDs (including U.2’s on mixed read/writes (something I assume is happening when using drives for SLOG and L2ARC)). The article suggested optane 900 and/or 905’s for mixed read/writes. what do you guys think?
This is just (mostly) for personal use… backing up desktops, backing up various homelab proxmox’s w/ versioning using Proxmox Backup Server pointing at a share, creating SMB share snapshots for versioning, plex, etc. I’m also trying to “future proof” as much as I can in terms of storage volume. Am I overdoing the resiliency? i.e. too many vdevs, or the fact that I’m doing z2 vs z1? or does it seem about right?
I’m also duplicating my very most important 1TB share offsite in s3 compatible storage and might spin up another TrueNas with either 1 or 2 sets of 2 mirrored 16TB drives to backup on site important shares… should give me 16 or 32TB of backup space minus overhead.
Any input is appreciated.