Complex problems with read speeds from raid 5 array in protools9.

[lupo]

Here is the issue I am facing: I have a high end workstation for PT9 with the 003 control surface. (Connected via fire wire). The C: drive of the system is a revodrive3 PCI-E SSD and the RAID array for storage is a 12TB RAID 5 running off of an Intel RT3WB080 RAID card. Each disk is 7200 RPM sata 3.
When I run benchmarks on the array I am getting read speeds of over 1600 MB/s at anything over 1024 byte blocks.
The array has MORE than enough speed to playback dozens of tracks without dropouts yet I still get this PT error when just trying a simple stereo 2 channel playback.
The thing about the array is that its read/write performance drops significantly if it is using a block size smaller than 1024. I think this may be contributing to my issue in that PT9 is reading/writing in blocks smaller than 1024.
Does anyone have any thoughts on how I can rectify this situation? Is there a way for me to go into the DAW properties and have protools read/write in larger blocks so i can get the full performance of the array? For those interested, here is a screenshot of my last benchmark of the array:
Intel Raid 5 Array
Any help or ideas you all may have would be greatly appreciated.
Thanks!
carl

 

[lupo]

Also, if it's an issue with block size, is there a way to force the OS to read/write in block sizes over 1024?

 

[lupo]

I checked the offset of the array using diskpart. Instead of seeing a normal size like 1024 it lists the offset at 129MB. This does not appear correct. Any thoughts? I could reinitialize the array if I had to by first moving the data off of it, but I'm not sure if that will help. Anyone with experience with the Intel RT3WB080 controller card? The revodrive3 SSD that is my C drive shows up as having a 110MB offset. Something seems off here.

 

[HobartTas]

Greetings

I don't know much about hardware raid but here are a couple of tips.

Firstly, can you change the stripe size and make it larger in the raid cards BIOS like on other raid cards?

Secondly, when you format the array then do you just let NTFS use the default cluster size of 4KB? in which case if its spread over say a 9 disk Raid 5 (8 data + 1 parity) then depending on the stripe size it may spread that 4KB into 512 byte blocks on each hard drive, for starters set the cluster size as high as you can go which I think is 64 KB and that would make 8KB go onto each of the 8 hard drives. With 64KB the only downside is you would have an average of 32KB wasted per file.

http://support.microsoft.com/kb/140365

If you don't have the large cluster size available as on option in the dropdown box when you format you may have to do it from a command prompt with admin privileges

format x: /fs:ntfs /a:64K

Thirdly, the 129MB offset sounds reasonable as when you install Win7 it puts in a 100MB hidden partition on the boot drive before it allocates the rest of the hard drive for your main partition. I gather it probably does the same for your data array, does disk manager show a 100MB partition? I'm guessing the other 29MB might be another extra hidden partition with the info about the raid array stored there. I would recreate the array starting at above the 129 MB mark, use diskpart and try to set the partition starting point at as high a power of 2 you can go (without losing too much space) like say 256MB which 2^19 512 byte sectors, this is 524288 decimal or 80000 hexadecimal exactly, you make have the problem like SSD's have with XP when it starts at sector 63 which is not at a 4K boundary.

Have fun, anyway that's my 2 cents worth

Cheers

 

[lupo]

I broke and rebuilt the entire array using 8k as default. It was set for 256k. It's still initializing so ill benchmark it later when it completes.