Hello,

 

i want to begin a project to go from my "multipurpose" unraid server from 2 unraid server :

• one dedicaded for storage, running 24/24, the most power efficient, in charge at most of down/uploading , manage media(audio/video), cold storage
• one for heavy lifting, running on request or/and on shedule, do task that are gpu/cpu-intensive, host dev tools (jenkins/git), host stream app, host a large database, ect ...

 

I have already two unraid key, a basic one and a plus, from an old project and my current NAS, which is not well configured and need a fresh start

I have a lot of question, that i will be posting in their categories,for parts and likes.

Here are my general questions :

1. For plex/jellifin app, will the storage on an other server cause lag ?
2. My two server will be on the same network i play with, is it worth to connect the two on their own network ?
3. I read that wake on lan work with unraid, did shutdonw on lan exist ?
4. I had the idea of using only SSD on the heavy one, will cache be useless ?
5. Which component is important for data transfer ? if i put a 10G network port in some "old" component will it be OK ?

comes donw to how you wnat to interact with it and what you want it to do.

Due to plex chanign plans and access i would recomend using jelly fin over plex unless you have a plex pass lifetime and user conecting to your plex also have a plex lifetime...

to answer your questiosn..

 

Quote

1 For plex/jellifin app, will the storage on an other server cause lag ?

2 My two server will be on the same network i play with, is it worth to connect the two on their own network ?

3 I read that wake on lan work with unraid, did shutdonw on lan exist ?

4 I had the idea of using only SSD on the heavy one, will cache be useless ?

5 Which component is important for data transfer ? if i put a 10G network port in some "old" component will it be OK ?

1. Plex/Jellyfin streaming from storage server — will there be lag?

Not if done right.
As long as:

Both servers are on gigabit or faster (ideally 10GbE) links

You're using direct file access protocols (NFS, SMB, or even iSCSI)

The storage server has enough I/O throughput (e.g., SSD cache, fast HDD pool)

Then Plex/Jellyfin on the "heavy" server can stream with no noticeable lag.
Transcoding performance, though, depends on CPU/GPU on the playback server, not where the files sit.

 

2. Separate network for inter-server communication — worth it?

Yes, especially if:

You're using 10G between them (e.g., via direct SFP+/RJ45)

You want to isolate storage traffic from internet/LAN access

You plan on heavy inter-server data movement (e.g., backups, syncing, large transfers)

A dedicated link (even 1G via a crossover cable or VLAN) helps reduce LAN congestion and increases transfer performance.

 

-Tailscale and sftp can be your friends here acorss the internet/intranet...

 

3. Shutdown on LAN (like Wake-on-LAN)?

-I'm personally not a fan of wol or sleep... if its 24/7 then its up 24/7 ... (I don't want snmp or magic packets on my network....)

There is plugins and other suport on the forum for this though...

 

Not natively standardized like WOL.
You'd need a workaround, such as:

SSH into the system and issue a poweroff

Use an API if the system supports one (some IPMI/iDRAC/iLO boards can do this)

Use Unraid's SSH plugin or user script for remote shutdown

Or integrate with something like Home Assistant to run shutdown commands via API or SSH

 

4. Only SSDs in heavy server — is cache drive still useful?

Maybe not needed.
If you're running all SSDs, Unraid's cache pool becomes redundant unless:

You're using a fast NVMe cache to buffer slower SATA SSDs

You want a dedicated scratch disk for certain apps/containers

Otherwise, just treat the SSD array as primary storage and disable Unraid's cache system.

 

I use unraid 7 to ditch the arrary and go pool only disk with zfs only. while more ram intensive, I don't have problems runnig unraid, nor hit a space and disk issues.
This said cache is recmoend to still have a btrfs voume for the swap plugin for zfs if ram resources are tight and other default setting that need chagned off the caceh disk...

--Some things like database shouldn't run on a array/zfs pool... thus it helpfull but no not needed..

 

5. What's important for fast data transfer (e.g., 10G on old hardware)?

NIC support is key: make sure the motherboard/CPU can handle the NIC's bandwidth.

PCIe lanes/bus speed must be sufficient — some old boards only have PCIe 2.0 x1 or x4, which bottlenecks 10GbE.

Disk speeds must match network speed — no point in 10GbE if your drives do 100MB/s.

(look into cmr/smr disk you want cmr for spinning rust for large storage...)

 

RAM and CPU can be a factor for handling network stacks and file services like SMB/NFS.

Use Intel or Mellanox 10G cards (better driver and Unraid support).

So yes, you can use 10G on older systems, but verify that PCIe slots and drivers won’t bottleneck it.

 

Thank you for your response.

For a database, where is right place to put it ? if not array/zfs pool, in a btrfs pool ?

28 minutes ago, Letypelouche said:

 

Thank you for your response.

For a database, where is right place to put it ? if not array/zfs pool, in a btrfs pool ?

Usually on a ssd / nvme this could be a single btrfs formatted cache pool.
Not all database are the same and you can add zfs meta/special disk like nvm to a zfs raidz1 to help and take over database like tasks...

as database need fast read and can have high writes...

that at least best practices... as plex has a sqlite database, it exists on my zfs raidz1.
Does it work, yes, but I have to acknowledge that heavy use will have higher wear adn tear on the disks.

Just things to keep note of.. as Hard drive, pc, technology is built to fail. usualy device have a MTF (mean time to fail) and disk last 3-5 years...

 

to not be reliant to a network i also recommend having the plex media local on the machine. While you can have the data on another. Being reliant on networks during maintenance, power outages, and issue can cascade and break things to easily... It better to run local and make backups staged to send backup back to the main.

 

Edited May 3, 20251 yr by bmartino1
typo - data