Introduction
I regularly meet with customers and discuss their Exchange Server architecture. Two topics frequently come up so I thought I'd discuss them with everybody and give my perspe...
Updated Jul 01, 2019
Version 2.0
Blog Post
Question about i/o performance.
Disk Bottleneck Detected
The Microsoft® Exchange Server Analyzer Tool has determined that your disk system is currently running within 20 percent of the expected maximum available throughput. This determination is made by one of the following calculations:
• Measuring disk latencies. The performance counters that indicate latency are LogicalDiskAvg.Disk sec/Read and LogicalDiskAvg. Disk sec/Write.
• Comparing the maximum possible disk I/O per second (IOPS) of your current disk configuration and spindle count against the current IOPS value, as recorded by Performance Monitor (Perfmon).
I need some clarification please.
Perfmon (or the OS for that matter) is not aware of disk configuration (i.e.raid groups) at the raid controller or disk array. That is, the OS kernel has no idea of what type of “physical” disk he sees, this “physical” disk could be a raid 10, raid 5, raid 3, or raid 50 on the back end and the os will see it as just one physical disk.
That said, I have a server attached to an array via FC, by different means I have confirmed that the disk utilization at the array is always almost nil (3 to 17%) however troubleshooter is complaining about disk performance. (I used different disk performance tools like IOmeter, and I have been able to load the back end disks to 100%)
By changing different parameters (io/interrupt ratio or NumFcpContext for example) I managed to increase disk utilization at the back end from 3-17% to 10-19%, but troubleshooter keeps complaining about disk performance.
Why doesn’t exchange use all the resources available to it?
How would you explain it?
TIA
Disk Bottleneck Detected
The Microsoft® Exchange Server Analyzer Tool has determined that your disk system is currently running within 20 percent of the expected maximum available throughput. This determination is made by one of the following calculations:
• Measuring disk latencies. The performance counters that indicate latency are LogicalDiskAvg.Disk sec/Read and LogicalDiskAvg. Disk sec/Write.
• Comparing the maximum possible disk I/O per second (IOPS) of your current disk configuration and spindle count against the current IOPS value, as recorded by Performance Monitor (Perfmon).
I need some clarification please.
Perfmon (or the OS for that matter) is not aware of disk configuration (i.e.raid groups) at the raid controller or disk array. That is, the OS kernel has no idea of what type of “physical” disk he sees, this “physical” disk could be a raid 10, raid 5, raid 3, or raid 50 on the back end and the os will see it as just one physical disk.
That said, I have a server attached to an array via FC, by different means I have confirmed that the disk utilization at the array is always almost nil (3 to 17%) however troubleshooter is complaining about disk performance. (I used different disk performance tools like IOmeter, and I have been able to load the back end disks to 100%)
By changing different parameters (io/interrupt ratio or NumFcpContext for example) I managed to increase disk utilization at the back end from 3-17% to 10-19%, but troubleshooter keeps complaining about disk performance.
Why doesn’t exchange use all the resources available to it?
How would you explain it?
TIA