The Windows NT Performance Monitor can help you isolate performance bottlenecks. Many Microsoft Exchange Server components create their own performance counters that provide more detailed information about performance bottlenecks. Using the Performance Monitor and the logging facility will help you identify problem areas. The main areas to consider when searching for bottlenecks are memory accesses, CPU usage, disk I/O, and network I/O.
The Performance Monitor counters for disk monitoring are turned off by default because they erode Windows NT Server performance by 1 to 2 percent. The Performance Monitor should be turned off when you are not specifically monitoring disk performance.
The Performance Monitor counters indicate the number and size of reads and writes that are being generated for a particular disk. They also show the percentage of time your application spends accessing a particular disk or disk strip set, as well as other information.
To turn on disk performance monitoring for a Windows NT Server computer, type diskperf -y at a command prompt and restart the computer. You may want to start the disk performance monitor on all your development servers before you begin testing your application.
Logical Disk
Memory
Processor
Redirector
Server
Note Before running a test, choose Chart from the View menu. Add basic counters such as:
Processor:%Processor Time
Logical Disk:C,D
This allows you to monitor the test until it is completed and the counters return to idle levels. Also, when you get all the settings, save your layout by choosing Save Workspace from the File menu. This will allow you to restore the settings at a later date.
To monitor a test in session, choose Chart from the View menu.
Processor:% Processor Time
Logical Disk:% Disk Time
Redirector:Bytes Total/sec
Memory:Pages/sec
You can draw your first conclusions from the data results. For example, if the Percent Processor Time is listed at 100 percent during most of the test, your application is CPU-bound.
For example, your test may have sent 100 messages. Multiply the Disk Bytes per second by the time interval as shown in Graph Time in the Chart View. Divide by 100 messages to determine the Disk I/O required per message.
Analyze the data obtained for each component to identify the direct source of any problems. Recommendations for isolating problems include the following:
For more detailed information about the Performance Monitor, see Optimizing Windows NT in the Windows NT Server Resource Kit.