During some tests I have performed on SQL Server on prefetch subject, I had to use SQLIO tool.
In this test, it is started with usual parameters
sqlio -kR -t20 -dG -s600 -b64
and interesting parameter is -t20. This parameter means that sqlio will run 20 threads.
In parallel with sqlio I started performance monitor for parameter Logical disk: Current Disk Queue Length. During sqlio work, Current Disk Queue Length was constantly 20, like number of threads.
What does these results mean? Does it mean that disk can process only one request at the time?
I have found some explanation regarding this parameter: "Current Disk Queue Length is the number of requests outstanding on the disk at the time the performance data is collected. Multi-spindle disk devices can have multiple requests that are active at one time, but other concurrent requests are awaiting service. This counter might reflect a transitory high or low queue length, but if there is a sustained load on the disk drive, it is likely that this will be consistently high. Requests experience delays proportional to the length of this queue minus the number of spindles on the disks. For good performance, this difference should average less than two."
So, this is maybe an answer to my question. I probably have only one spindles. And Current Disk Queue Length is equal to number of requests.
Another interesting thing is difference between disk from the point of sequential and random write.
Here are the commands and results:
Disk A
sqlio -kW -dC -s10 -b64 -fsequential
throughput metrics:
IOs/sec: 1535.20
MBs/sec: 95.95
sqlio -kW -dC -s10 -b64 -frandom
throughput metrics:
IOs/sec: 120.60
MBs/sec: 7.53
Disk B
sqlio -kW -dH -s10 -b64 -fsequential
throughput metrics:
IOs/sec: 102.90
MBs/sec: 6.43
sqlio -kW -dH -s10 -b64 -frandom
throughput metrics:
IOs/sec: 126.10
MBs/sec: 7.88
It can be seen that Disk A comparing to Disk B has about 15 times faster sequential write, while random write is similar on both disk. What can be concluded from these figures?
Very interesting and informative link describing SQLIO usage, can be found here
In this test, it is started with usual parameters
sqlio -kR -t20 -dG -s600 -b64
and interesting parameter is -t20. This parameter means that sqlio will run 20 threads.
In parallel with sqlio I started performance monitor for parameter Logical disk: Current Disk Queue Length. During sqlio work, Current Disk Queue Length was constantly 20, like number of threads.
What does these results mean? Does it mean that disk can process only one request at the time?
I have found some explanation regarding this parameter: "Current Disk Queue Length is the number of requests outstanding on the disk at the time the performance data is collected. Multi-spindle disk devices can have multiple requests that are active at one time, but other concurrent requests are awaiting service. This counter might reflect a transitory high or low queue length, but if there is a sustained load on the disk drive, it is likely that this will be consistently high. Requests experience delays proportional to the length of this queue minus the number of spindles on the disks. For good performance, this difference should average less than two."
So, this is maybe an answer to my question. I probably have only one spindles. And Current Disk Queue Length is equal to number of requests.
Another interesting thing is difference between disk from the point of sequential and random write.
Here are the commands and results:
Disk A
sqlio -kW -dC -s10 -b64 -fsequential
throughput metrics:
IOs/sec: 1535.20
MBs/sec: 95.95
sqlio -kW -dC -s10 -b64 -frandom
throughput metrics:
IOs/sec: 120.60
MBs/sec: 7.53
Disk B
sqlio -kW -dH -s10 -b64 -fsequential
throughput metrics:
IOs/sec: 102.90
MBs/sec: 6.43
sqlio -kW -dH -s10 -b64 -frandom
throughput metrics:
IOs/sec: 126.10
MBs/sec: 7.88
It can be seen that Disk A comparing to Disk B has about 15 times faster sequential write, while random write is similar on both disk. What can be concluded from these figures?
Very interesting and informative link describing SQLIO usage, can be found here
No comments:
Post a Comment