JBHAIRE2004
What is this storage?
- Joined
- Feb 7, 2011
- Messages
- 2
I have a prior generation SAN unit and I am trying to tune my subsystem for maximized, sustainable, Read \Write rates. I have been doing lots of reading and trial and error testing. There is so much conflicting information that I have managed to confuse myself. I’m not accomplishing anything at this point so I think it time to step back and ask for help. How would one best go about tuning storage for maximized Read \ Write rates?
My intent is to be maximizing Read \ Write rates on the SAN for large file storage and virtualization. The majority of the static files on the SAN will be 700MB ISO Images, 1-2GB video files, 4GB DVD ISO Images and 4GB Video files. I am also going to be putting Hyper-V virtual machines on the SAN. The VM VHDs will be a combination of dynamically expanding VHDs and fixed size VHDs.
Details and specs about the hardware in play:
- (16) x 1TB WD RE3 SATA Drives
- Infortrend A16F-R2221 FC to SATA SAN with Redundant controllers.
- Controllers are only SATA-I for all intents and purposes
- For max compatibility with the controller, I jumper the WD RE3 drive to hard set them at 1.5GHz
- 2GB sitck of RAM for cache in each controller
(Controller only uses 31-33% of cache or 630MB for Read \ Writes)
- Redundant 2GB Fibre channel ports to each controller on the SAN and to the Server HBA
- Brocade Silkwork 3902 32 Port 2GB FC Switch
- I am focused on Maximized Read \ Write rates for sequential reads and writes
(Tell me is that is wrong for my intended purposes).
- Separating the (16) Drives into (2)x 8 Drive RAID-5 LD Arrays
o Array A assigned to controller A
o Array B assigned to controller B.
- SAN Raid Level Stripe Sizes Available:
o 16K (16,384 bytes)
o 32K (32,768 bytes)
o 64K (65,536 bytes)
o 128K (131,072 bytes)
o 256K (262,144 bytes)
o 512K (524,288 bytes)
o 1024K (1,048,576 bytes)
- Windows Partition Sector Sizes Available:
o 512 - 512 bytes (.5K)
o 1024 – 1,024 bytes (1K)
o 2048 – 2,048 bytes (2K)
o 4096 – 4,096 bytes (4K)
o 8192 – 8,192 bytes (8K)
o 16K - 16,384 bytes (16K)
o 32K - 32,768 bytes (32K)
o 64K - 65,536 bytes (64K)
- Currently only 1 Partition Per LD on the SAN and in Windows.
- Server’s OS is Windows Server 2008 R2 x64 with RC SP1
- (If it matters Current server is Dell PE860 – Socket 775 2.4Ghz Quad Core with 8GB RAM
Once ready will connect SAN to my Dell PE1950 GIII –Socket 771 (2)x 2.33Ghz Quad Cores with 48GB RAM)
- (Multipath I/O Feature / MPIO driver are installed on DE860 if it could be effecting performance)
I have been using bst514 (aka Bart’s Stuff Test v5.1.4) and ATTO Disk Benchmark to conduct read and write tests. (See attached screenshots for reference). I always copy the application executable local to the drive that I am test to keep the results as apples to apples as possible.
What I have been attempting to use those apps to determine are the combination of SAN Raid Level Stripe Size and NTFS Windows Partition Sector Size that yield the highest possible read write rate.
Where things got fuzzy for me, was in trying to understand the bench mark results. Each application (bst514 and ATTO Disk Benchmark) had a setting that threw me when I examined things closer.
In bst514, under options there is a block size setting. Changing the block size affects the Read \ Write rates. While I run the test I monitor the writes from the GUI console of the SAN and see supporting Read \ Write rates reflected there as less as the Cache utilization.
In ATTO Disk Benchmark its the “Transfer Size” and “Total Length” although it also bring in I/O and I am not so familiar with the nuances of I/O. Again while I test I watch the SAN GUI and Read \ Write rates reflected in the app appear fairly supported.
By adjusting up the block size and transfer size within the test parameters of the applications; I was able to make what appeared to be a positive impact on the Read \ Write rates. However in an actual file copying test, from a windows drive to the drive I was testing. I didn’t see those highest speeds. How does Windows writes in the real world equate to these tests?
Any help or guidance to help put me on the right track would be greatly appreciated.
My intent is to be maximizing Read \ Write rates on the SAN for large file storage and virtualization. The majority of the static files on the SAN will be 700MB ISO Images, 1-2GB video files, 4GB DVD ISO Images and 4GB Video files. I am also going to be putting Hyper-V virtual machines on the SAN. The VM VHDs will be a combination of dynamically expanding VHDs and fixed size VHDs.
Details and specs about the hardware in play:
- (16) x 1TB WD RE3 SATA Drives
- Infortrend A16F-R2221 FC to SATA SAN with Redundant controllers.
- Controllers are only SATA-I for all intents and purposes
- For max compatibility with the controller, I jumper the WD RE3 drive to hard set them at 1.5GHz
- 2GB sitck of RAM for cache in each controller
(Controller only uses 31-33% of cache or 630MB for Read \ Writes)
- Redundant 2GB Fibre channel ports to each controller on the SAN and to the Server HBA
- Brocade Silkwork 3902 32 Port 2GB FC Switch
- I am focused on Maximized Read \ Write rates for sequential reads and writes
(Tell me is that is wrong for my intended purposes).
- Separating the (16) Drives into (2)x 8 Drive RAID-5 LD Arrays
o Array A assigned to controller A
o Array B assigned to controller B.
- SAN Raid Level Stripe Sizes Available:
o 16K (16,384 bytes)
o 32K (32,768 bytes)
o 64K (65,536 bytes)
o 128K (131,072 bytes)
o 256K (262,144 bytes)
o 512K (524,288 bytes)
o 1024K (1,048,576 bytes)
- Windows Partition Sector Sizes Available:
o 512 - 512 bytes (.5K)
o 1024 – 1,024 bytes (1K)
o 2048 – 2,048 bytes (2K)
o 4096 – 4,096 bytes (4K)
o 8192 – 8,192 bytes (8K)
o 16K - 16,384 bytes (16K)
o 32K - 32,768 bytes (32K)
o 64K - 65,536 bytes (64K)
- Currently only 1 Partition Per LD on the SAN and in Windows.
- Server’s OS is Windows Server 2008 R2 x64 with RC SP1
- (If it matters Current server is Dell PE860 – Socket 775 2.4Ghz Quad Core with 8GB RAM
Once ready will connect SAN to my Dell PE1950 GIII –Socket 771 (2)x 2.33Ghz Quad Cores with 48GB RAM)
- (Multipath I/O Feature / MPIO driver are installed on DE860 if it could be effecting performance)
I have been using bst514 (aka Bart’s Stuff Test v5.1.4) and ATTO Disk Benchmark to conduct read and write tests. (See attached screenshots for reference). I always copy the application executable local to the drive that I am test to keep the results as apples to apples as possible.
What I have been attempting to use those apps to determine are the combination of SAN Raid Level Stripe Size and NTFS Windows Partition Sector Size that yield the highest possible read write rate.
Where things got fuzzy for me, was in trying to understand the bench mark results. Each application (bst514 and ATTO Disk Benchmark) had a setting that threw me when I examined things closer.
In bst514, under options there is a block size setting. Changing the block size affects the Read \ Write rates. While I run the test I monitor the writes from the GUI console of the SAN and see supporting Read \ Write rates reflected there as less as the Cache utilization.
In ATTO Disk Benchmark its the “Transfer Size” and “Total Length” although it also bring in I/O and I am not so familiar with the nuances of I/O. Again while I test I watch the SAN GUI and Read \ Write rates reflected in the app appear fairly supported.
By adjusting up the block size and transfer size within the test parameters of the applications; I was able to make what appeared to be a positive impact on the Read \ Write rates. However in an actual file copying test, from a windows drive to the drive I was testing. I didn’t see those highest speeds. How does Windows writes in the real world equate to these tests?
Any help or guidance to help put me on the right track would be greatly appreciated.
