Frozen Fractals
Cadet
- Joined
- Jul 20, 2023
- Messages
- 1
Motherboard make and model: Lenovo 03T8873 (TS140)
CPU make and model: Intel Xeon E3-1246v3
RAM quantity: 32GB DDR3-1600 ECC
Hard drives, quantity, model numbers, and RAID configuration, including boot drives: (10) HGST He10 in single RAIDz2 vdev for main data store, (1) Samsung 850 Evo 500GB for OS boot, (2) Intel S3710 400GB which may be used for L2ARC and/or SLOG
Hard disk controllers: LSI 9211-8i IT and Intel C226
Network cards: Mellanox MCX311A-XCAT
Having read up a bit on ZIL/SLOG and how its used within ZFS, my understanding is that it's really only useful for synchronous writes and not at all for async writes (which is most writes via NFS/SMB). However, I have a quick question for my specific use case.
My use cases for my NAS is mainly for general purpose file and media via NFS. However, I also plan on using the array for backups: Veeam for vSphere VMs and physical Windows/Linux hosts and Time Machine for macOS hosts. I'll have two datasets on this single vdev: one for my unstructured data (media, etc.) and one for my Veeam and Time Machine backup repositories. I'll be using NFS for the unstructured data and Veeam repo, and SMB for the Time Machine repo.
To my understanding, Veeam and Time Machine both use sync writes vs. async (at least from the reading I've done). Would it benefit me to use one of my S3710 SSDs for SLOG to accelerate my backup jobs or will it not make sense in this case, since the lion's share of my data will be unstructured data that'll be all async? It'd be very clear to me if I had separate vdevs for unstructured data and backup data, but since I'll be putting it all on one vdev that's where the question comes from.
Also, is it best practice to mirror SLOG? I understand the only time you'll lose data in flight (stored in ZIL in SLOG and RAM) is if you have a system failure and SLOG drive failure simultaneously?
Thanks!
CPU make and model: Intel Xeon E3-1246v3
RAM quantity: 32GB DDR3-1600 ECC
Hard drives, quantity, model numbers, and RAID configuration, including boot drives: (10) HGST He10 in single RAIDz2 vdev for main data store, (1) Samsung 850 Evo 500GB for OS boot, (2) Intel S3710 400GB which may be used for L2ARC and/or SLOG
Hard disk controllers: LSI 9211-8i IT and Intel C226
Network cards: Mellanox MCX311A-XCAT
Having read up a bit on ZIL/SLOG and how its used within ZFS, my understanding is that it's really only useful for synchronous writes and not at all for async writes (which is most writes via NFS/SMB). However, I have a quick question for my specific use case.
My use cases for my NAS is mainly for general purpose file and media via NFS. However, I also plan on using the array for backups: Veeam for vSphere VMs and physical Windows/Linux hosts and Time Machine for macOS hosts. I'll have two datasets on this single vdev: one for my unstructured data (media, etc.) and one for my Veeam and Time Machine backup repositories. I'll be using NFS for the unstructured data and Veeam repo, and SMB for the Time Machine repo.
To my understanding, Veeam and Time Machine both use sync writes vs. async (at least from the reading I've done). Would it benefit me to use one of my S3710 SSDs for SLOG to accelerate my backup jobs or will it not make sense in this case, since the lion's share of my data will be unstructured data that'll be all async? It'd be very clear to me if I had separate vdevs for unstructured data and backup data, but since I'll be putting it all on one vdev that's where the question comes from.
Also, is it best practice to mirror SLOG? I understand the only time you'll lose data in flight (stored in ZIL in SLOG and RAM) is if you have a system failure and SLOG drive failure simultaneously?
Thanks!
