I just wanted to thank you for posting this thread. I spent a day or so sorting out tier configurations with a six disk array of 4TB Hitachi 7200 drives, and two SSDS. The performance on larger high capacity writes was not good using storage spaces with 2 x Intel 530 SSD tier and 5 column parity over six drives, when really pushing from a 10G workstation pushing a 500MB/s backup. I ended up going back to RAID 5 using a rocketraid 2720. With 16GB of RAM in the windows 2012 server, these files are being sucked in at 1000MB/s. I'll post up my powershell script here for reference. Although the Powershell interface is a pretty crazy way to set and monitor an array (sheesh!) I did appreciate being able to tear the array down, and simply past the entire script into the powershell window to do an entire new setup in a few seconds.
I will mention a few things that may help in any further testing you do based on a few comments you made.
1. RamDrives on both test machines are pretty much required to really test IO over 10G. RAMdisk software also will affect your speed. I found SoftPerfect Ramdisk to be one of the faster free packages. IMdisk is also free, but skewed top end results by nearly 200MB/s
2. Parkdale (a rather simple desk benchmark) does work over networked drives (you can map to shared ramdisk drives as well).
3. ATTO will work over networked drives if you use the SUBST command at the command prompt. Example SUBST m: \\192.168.0.1\raid_share_name
I'll add in a few screen grabs during my own quest for sustained 1000MB/s performance.
This is an example of multichannel (2 x 10G) Intel x540 connections with tuned drivers, jumbo frames enabled. Both workstations using Windows 8.1, so therefore SMB3
This is ATTO "fooled" by a SUBST command so is measuring array performance on the "server" from the workstation...both windows 8.1 Pro.
Cheers, Dennis (Cinevate - Tools for Filmmakers and Photographers)
I will mention a few things that may help in any further testing you do based on a few comments you made.
1. RamDrives on both test machines are pretty much required to really test IO over 10G. RAMdisk software also will affect your speed. I found SoftPerfect Ramdisk to be one of the faster free packages. IMdisk is also free, but skewed top end results by nearly 200MB/s
2. Parkdale (a rather simple desk benchmark) does work over networked drives (you can map to shared ramdisk drives as well).
3. ATTO will work over networked drives if you use the SUBST command at the command prompt. Example SUBST m: \\192.168.0.1\raid_share_name
I'll add in a few screen grabs during my own quest for sustained 1000MB/s performance.
This is an example of multichannel (2 x 10G) Intel x540 connections with tuned drivers, jumbo frames enabled. Both workstations using Windows 8.1, so therefore SMB3
This is ATTO "fooled" by a SUBST command so is measuring array performance on the "server" from the workstation...both windows 8.1 Pro.
Cheers, Dennis (Cinevate - Tools for Filmmakers and Photographers)