QA My Build

[dicecca112]

I'm in the process of building out a napp-It (OmniOS as the Base OS) stand alone box for serving media for Plex and General File Storage.

Here is the build sheet

6X HGST 6TB He Drives

64GB DDR4 ECC

SuperMicro X10SDV-4C-TLN4F Motherboard

LSI 9211-8i HBA

Supermicro SSD-DM016-PHI 16GB SATA DOM (Disk on Module)

Fractal Node 304 Case

OCZ RD400/400A M.2 SSD (Not Sure if it can be used)

Intel X520-DA2 10G Network Card for Direct Connect to Plex VM

64GB SuperMicro SuperDOM

My Synology NAS could hit a sustained throughput of 100MB/s but I'd like to hit more if possible. My thoughts was direct connect the NAS via the X520 Card in my Vmware Server to my Plex and Ripping VM via the 10G on the SuperMicro Board

Am I correct to think the best performance would be 3 Mirrored VDEVS (2X6TB) for a total of ~18TB of Storage?

My other question is SLOG, is one drive sufficient or can I mirror them? If mirroring is not necessary or possible, what size should I consider, and is there a rule of thumb for the size? Is it still 1/8 of your memory? With 64GB of memory I wouldn't need a separate L2ARC?

 

[whitey]

Build looks fairly solid to me, stripped/mirror vdev's (raid-10) for roughly 16-17TB usable (accounting for overhead), for SLOG use a Hitachi/HGST hussl 100GB or Intel DC s3700/s3710 200GB model. Should cost between $50-100 depending on deals you can find...worth every penny and then some. With 64GB of memory I would forego L2ARC personally.

 

[gea]

Everything is ok for ZFS
But why do you want an Slog for a ZFS filer for Plex?
ZFS is always consistent. You need the Slog only in cased where you can not allow to loose the data in the write cache. This is mainly true with old filesystems (VM storage) or transactional databases.

btw.
With 64 GB I would also skip the L2Arc, mostly not in use and much slower than the rambased Arc cache.

 

[dicecca112]

I mostly want the slog for future proofing in case I decide to do something that needs it in the future. I might eventually once cost of spinning storage comes down put my VMware VMs on it.

 

[gea]

about an Slog

If an slog fails, ZFS reverts to the onpool ZIL. You can export/ import pools with a missing slog. You mainly need an Slog mirror if you want to be save that an slog failure does not affect sync write performance. The risk of an slog failure with a crash at the same time that can lead to a dataloss of last writes is very minimal.

As a rule of thumb, the size of an slog should be at least the data that can be written in 7s. With 10G = around 120MB/s this means a minimal size of around 8GB. This is why a ZeusRAM (one of the best Slog devices) comes with 8GB.

For a good slog powerloss protection, low latency and continous high write iops in queue depth=1 is essential. A good slog may be an Intel S3700 SSD or NVMes from the Intel P-series.