Maximum array size per CPU? (Solved)

[Xcage]

Hi all,

 

question is pretty simple , but I want to be sure that I am not missing something.

 

what would be the maximum size for the array? I mean before I get hit on performance, of course not counting VMs or anything.

something like a zerotier/tailscale couple of other small dockers and a big array.

Currently planning on 20 16tb mechanical drives, 2 2tb NVME for cache, maybe some ssds for frequently changed data.

what would be by CPU needs? Ram needs etc 

 

I do have 5 unraid servers, but all of those aren’t exceeding 10tb each

 

exclude networking , got that figured out

Edited November 22, 2021 by Xcage

[itimpi]

For basic file serving capabilities any modern CPU should be fine and 4GB of RAM.   If you start running docker container and/or VMs then their requirements need to be added.

[Xcage]

15 minutes ago, itimpi said:

For basic file serving capabilities any modern CPU should be fine and 4GB of RAM.   If you start running docker container and/or VMs then their requirements need to be added.

So I could have maximum network load of whatever my network is (up to 10gbit) and any modern Cpu will handle it? Say 11th gen at least 4 cores

[itimpi]

57 minutes ago, Xcage said:

So I could have maximum network load of whatever my network is (up to 10gbit) and any modern Cpu will handle it? Say 11th gen at least 4 cores

I would expect that to be overkill in CPU terms.   On most Unraid systems the drives and/or disk controllers are what limits performance

[Ford Prefect]

22 hours ago, Xcage said:

So I could have maximum network load of whatever my network is (up to 10gbit) and any modern Cpu will handle it? Say 11th gen at least 4 cores

I am running an i3-8100 and it does handle a 10Gbps link fine. Remember that SMB is single threaded and that both sides would have to be equally potent.

Hence, for unraid, you will never, ever achieve this without a PCIe based NVMe (cache) drive...as the performance of the array is maxed at that of a single disk.

Since a SATA HDD or SSD is simply not fast enough.

[sdamaged]

imo, 10Gb networking isn't worth it on unRAID.  You're limited by the speed of a single drive. For that many disks and speed seemingly important to you, i would be looking at a different solution personally.  If sticking with unRAID, then it's madness to purchase all 20 drives in one go, unless you already have a ton of data to copy over and use the space.  If not, then the best thing is to purchase a couple (of mechanical drives), and when they are approaching being full, purchase more at what will likely be a lower price (and possibly higher capacity drives).  This is one of unRAID's main strengths and what it excels at, along with an excellent and user friendly docker experience

Edited November 3, 2021 by sdamaged

[hawihoney]

4 hours ago, sdamaged said:

10Gb networking isn't worth it on unRAID

 

It depends. PCIe connected NVMe drives handle >3000 MB/s quite easily. So Cache/Pool-Writes would utilize networking at that speed.

 

[sdamaged]

On 11/3/2021 at 7:03 PM, hawihoney said:

 

It depends. PCIe connected NVMe drives handle >3000 MB/s quite easily. So Cache/Pool-Writes would utilize networking at that speed.

 

Correct, assuming you're copying from the cache drive before the mover has ran or copying to the array for the first time and hitting the cache disk. 

Edited November 5, 2021 by sdamaged

[hawihoney]

On 11/5/2021 at 11:49 AM, sdamaged said:

assuming you're copying from the cache drive before the mover has ran or copying to the array for the first time and hitting the cache disk

 

With a proper design/layout of the usage of system resources that's no problem. That's what I wrote exactly: "Cache/Pool-Writes". That does include Cache/Pool-Reads as well.

 

Edited November 22, 2021 by hawihoney