Should I choose all flash nas?

[leonfox28]

Hi guys,

  My group is trying to build a new NAS for following needs:

  1. 10GBE network;

  2. iscsi for esxi to support about 10 to 15 people working on there own virtual desktop;

  3. nfs for those virtual machines;

  4. For our daily work, it's mostly will be coding, PCB layout, simulation and etc. They are almost all small files so maybe the random 4K performance is most important?

  We've decided to choose TrueNas Scale for host OS because our limited budget but still not sure what we should choose for storage devices: All U.2 NVMe SSD or HDD data vdev + SSD special vdev for metadata? If we choose the second choice, should we add slog devices?

Thanks a lot!

  

[RollinLower]

what are the rest of the system specs you are working with? TrueNAS can allocate a lot of ZFS caching to memory if you have enough RAM, so adding more caching or slog devices might not be needed.

[jaslion]

Wait so this isnt just a nas but a virtual desktop host?

 

For 15 people coding, simming,...? I dont think a powerfull enough cpu exists on the planet to do this

[leonfox28]

4 minutes ago, RollinLower said:

what are the rest of the system specs you are working with? TrueNAS can allocate a lot of ZFS caching to memory if you have enough RAM, so adding more caching or slog devices might not be needed.

single epyc 7543 CPU, 64x8 Reg Ecc RAM, BCM57412 for RDMA support

[leonfox28]

5 minutes ago, jaslion said:

Wait so this isnt just a nas but a virtual desktop host?

 

For 15 people coding, simming,...? I dont think a powerfull enough cpu exists on the planet to do this

nonono, still a dedicate NAS, using iscsi and nfs to attach storage to another server running esxi.

[Bersella AI]

Yes, it is always better to prepare all storage with SSDs in regard of performance. I was told by an online guideline that more RAM as cache and the Second-Layer Adaptive Replacement Cache (L2ARC) could help in concurrent reads, and the Separated ZFS intend Log (SLOG) could help in concurrent writes, from both of which iSCSI & NFS would be benefited. The Special Metadata Vdev could also make more sense in small transactions. However, they may not become that useful in specific scenarios, and you have to monitor the actual workload on the server to tweak its performance more wisely.

I once attempted to make a SLOG with an Optane SSD in my home server, and found it having negligible throughput in the monitoring panel -- nearly all of my home traffic relies on HTTP & SMB, no iSCSI or other concurrent stuff.