Just curious on performance thoughts comparing system configs and actual results.  If one had a 3-4yr old xeon (say 2.6-3.4 ghz) processor 6 core, compared to a 3-4 yr old i5 or i7, would unraid (all other things being equal) see a significant performance boost?  I'm thinking it should (cost effectively is another question) as all channels on a Xeon-based board (for the most part) should be better (wider PCI...handle more disks, etc...)?

 

Or would it be smarter to spend the $$ on high-quality HBA's and faster drives?

 

I'm currently getting 113mb/s (using SMB) copy PC to unRaid (SSD cache) and about the same copy from Cache to PC.  Copy direct from Disk (i.e. using Disk 5 share to bypass cache) I get about the same (102-108 MB/s).  All via gigabit ethernet (cat 6 cable) on my rig in my signature.  I've got 10 drives plus parity on AOC-SAS2LP-MV8 (PCI X16 slot), Rest on the Motherboard except for Cache (1 drive on PCI X8 slot).

 

So I'm thinking that in my rig, the bottleneck is the drives (in other words, not much I can do to improve performance).  DMI 2.0 (Z77 chipset) supports 2-4 GB/second concurrent transfer rates via two unidirectional lanes.  2GB/s in one direction.  So...(not sure my math is correct, and that's some of the feedback I'm looking for)... 

 

Hard Drives - I have 10 data drives in my system, which are using the following:

 

• AOC-SAS2LP-MV8 - 8 Drives via PCI 2.0 X16 Slot (8 / 2GB/s = 250 MB/S max through that channel?
   
  
• MB SATA Ports - 2 Drives via direct SATA i/f to Platform Controller Hub = 600mb/s or 300mb/s max through those channels.  My parity drive is on the 6GB/s channel, 5TB ST5000DM (maxes at 150-170 MB/s).
   
  

 

Cache Drive - I have 1 cache drive which uses the following:

• SYBA SSD Caddy - 1 drive via PCI 2.0 X8 Slot = I assume Shared Access with the 8 Disks on the AOC-SAS2LP-MV8?  So 9 / 2GB/s = 222 mb/s.  The drive supports up to 285 MB/s. 
   
  
• 
  

 

So, in my system, it appears that the mechanical drives are the bottleneck (inside the server)?  I think that the PCI channel is only split by devices doing an active transfer?

 

I'm thinking that I should probably max out the motherboard Sata before I go to PCI channels?

 

For file transfer, gigabit ethernet should do 125MB/s, I'm getting 113MB/s.  Not bad, perhaps parity checks or some other bit is consuming the difference?

 

Is all this making sense?

 

 

If you want to go over the ~100 range you need faster network.

 

Going back to CPU choices, check the Passmark scores for a closer to definitive answer.