Oracle Database Appliance IO performance was measured with different tools: Orion, Calibrate, oakcli.
Here is summary:
Despite of ORION documentation examples showed against raw (character) devices I used block devices (first column) because Linux deprecated raw devices.
SAS Disks were tested with 8K block size.
SSD columns:
Because of 512b redo log block size I wanted to measure SSD performance for 512 bytes blocks. But ORION don't have the 512 bytes block size. The 1K is minimum.The steps are:
- 1K - because it nearest to the 512b
- 4K - because it is default SSD block size, so this size is optimal for IO operations
- 8K - for comparison purposes. If you put db file on SSD.
Calibrate IO:
I thinks that number of IOPS=14238 is double measured mistake .
The more real number is an half of this value 7119 ( theoretical value is: 20 disks * 320 IOPS = 6400 IOPS) . Calibrate_IO uses database block size, the 8K in our case.
Here is summary:
Despite of ORION documentation examples showed against raw (character) devices I used block devices (first column) because Linux deprecated raw devices.
SAS Disks were tested with 8K block size.
SSD columns:
Because of 512b redo log block size I wanted to measure SSD performance for 512 bytes blocks. But ORION don't have the 512 bytes block size. The 1K is minimum.The steps are:
- 1K - because it nearest to the 512b
- 4K - because it is default SSD block size, so this size is optimal for IO operations
- 8K - for comparison purposes. If you put db file on SSD.
Calibrate IO:
I thinks that number of IOPS=14238 is double measured mistake .
The more real number is an half of this value 7119 ( theoretical value is: 20 disks * 320 IOPS = 6400 IOPS) . Calibrate_IO uses database block size, the 8K in our case.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.