Usually the issue with poorly performed VMs that do not have enough IOPS and IOPS available is less than 30. Best practice per VM you need is:
Good IOPS = 30 IOPS per VM
Disk RPM / IOPS available
7,200 RPM = 100 IOPS
10,000 RPM = 150 IOPS
15,000 RPM = 230 IOPS
Calculate total IOPS per VM
- Find the speed of the drive and determine the IOPS.
- Each drive in the RAID volume within the SAN contributes to the total IO available for that volume.
- IOPS per VM is calculated by dividing the total capacity available by the number of VM running on that volume.
For example, if you add eight 15,000 RPM drives, the total IOPS available in that volume is 230 x 8 = 1840. If you had 80 VMs running on LUNs located on this RAID volume, each VM gets an average of 1840 / 80 = 23 IOPS per VM.
In this example, to obtain acceptable performance (depending on the application), it should not exceed 1840 / 30 = 61 VMs on that volume.
Calculate functional IOPS
RAID has write IO penalty, depending on what RAID set.
RAID 0 = 0
RAID 1 = 2
RAID 5 = 4
RAID 6 = 6
RAID 10 = 2
Total RAW IOPS = disk speed IOPS * number of disks
Total FUNCTIONAL IOPS = ((Total RAW IOPS * Write %) / (RAID PENALTY)) + (Total Raw IOPS * Read%)
In our example ,
Total Raw IOPS = 175*8 = 1400 IOPS ( Since 15K RPM disk can give 175 IOPS)
When on RAID-5,
Scenario 1(80%Write20%Read) Functional IOPS = (((1400*0.8))/(4))+(1400*0.2) = 560 IOPS
Scenario 2(20%Write80%Read) Functional IOPS = (((1400*0.2))/(4))+(1400*0.8) = 1190 IOPS
When on RAID-1,
Scenario 1(80%Write20%Read) Functional IOPS = (((1400*0.8))/(2))+(1400*0.2) = 840 IOPS
Scenario 2(20%Write80%Read) Functional IOPS = (((1400*0.2))/(2))+(1400*0.8) = 1260 IOPS
Measuring storage performance
- IOPS (or commands/sec)
- Response time (latency)
- Throughput (bandwidth, megabytes/sec)
Other consideration
Backing up the storage consumes more IOPS and adds an additional strain on the volume. In this case you may want to have an additional overhead capacity to handle this. To determine the amount of additional capacity you may require to perform backups, contact backup vendor.
Larger drives generally equate lower performance.
More spindles generally equals greater performance.
Reference
http://sudrsn.wordpress.com/2010/12/25/iops-raid-penalty-and-workload-characterization/
