We've been testing Microsoft's famous solution ; storage pools , SaS and finally CSV for a cluster environment IOPS .
we have SaS disks connected through JBODs and SaS HBA cards .
before making a CSV ; I used SQLIO and got about 150,000 IOPS which is acceptable ( ReadONLY , 4k , Random )
After making the CSV volume , on the coordinator node ; I am getting : 100,000 ( 50,000 less right away )
on a noncorordinator node ; I am getting only 25000 IOPS .
IF I change the owner (and make it coordinator) , then I get 100,000 IOPS again .
This is so much difference . we don't get actual IOPS that we paid for . we get 25 percent of it . any thoughts ?
we have SaS disks connected through JBODs and SaS HBA cards .
before making a CSV ; I used SQLIO and got about 150,000 IOPS which is acceptable ( ReadONLY , 4k , Random )
After making the CSV volume , on the coordinator node ; I am getting : 100,000 ( 50,000 less right away )
on a noncorordinator node ; I am getting only 25000 IOPS .
IF I change the owner (and make it coordinator) , then I get 100,000 IOPS again .
This is so much difference . we don't get actual IOPS that we paid for . we get 25 percent of it . any thoughts ?