When you first start using Performance Monitor, the number of performance counters might seem overwhelming. It's not necessary to be familiar with all of the performance counters. Some are appropriate only for programmers writing Windows NT–based applications; others are useful for vendors who need to test hardware performance. Later chapters in this section recommend certain counters to diagnose problems on each component of your computer.
Begin by logging the Logical Disk, Memory, Process, Processor, System, and Thread objects. Run the log for at least a few days at an Update Interval of 60 seconds. Then, chart the results.
Create a chart settings file with counters that give you a broad view of your system. The default counters are a good starting point. When you open the Add To box, the Processor: % Processor Time counter is selected because this counter is used most often. Each object has a default counter which is highlighted when you select the object. These counters were selected as defaults because they are excellent indicators of the object's activity.
The following table shows the default counter for the most commonly monitored objects.
Object | Default Counter | Description | |
Cache | Data Map Hits % | How often requested data is found in the cache. This is an indicator of application I/O. A poor cache hit rate may indicate a memory shortage. | |
Logical Disk | Avg. Disk Queue Length | A measure of the activity of each logical partition of the disk. An Avg. Disk Queue Length of 1.0 indicates that the logical disk was busy for the entire sample time. Busy time includes all processing time for a disk I/O request, including driver time and time in the queue, so values for a single logical disk may exceed 1.0. Sustained high values over time indicated a possible disk bottleneck. | |
Memory | Pages/sec | The number of pages between main memory and the disk drives in each second. If this counter is consistently high, memory is in short supply. Sustained paging degrades performance. | |
Objects | Processes | An instantaneous count of the number of processes running. When charted with Processor: % Processor Time, it shows the effect on the processor of adding and removing processes. | |
Physical Disk | Avg. Disk Queue Length | A measure of the activity of the disk subsystem. It is the sum of Avg. Disk Queue Length for all logical partitions of the disk. This is a good measure of disk activity when measuring multiple physical disks in a disk set. | |
Process | % Processor Time | A measure of each process's use of the processor. | |
Processor | % Processor Time | A good indicator of the demand for and efficiency of a processor. | |
System | % Total Processor Time | Include this counter to monitor multiprocessor systems. It combines the average processor usage of all processors into a single counter. | |
Thread | % Processor Time | Threads are the components of a process that execute its code on the processor. This counter indicates which threads are getting processor time. | |
You should also include counters to monitor network throughput. The counters you choose depend upon your network protocol and whether the computer is primarily a client, a server, or both. NetBIOS: Bytes Total/sec for NWLink or Network Interface: Bytes Total/sec for TCPIP/SNMP are good overview counters.
If the computer is primarily a server, include Server: Bytes Total/sec to monitor network activity. You might also want to include Server: Context Blocks Queued/sec and System: Total Interrupts/sec.
You should also include a few alerts in the overview settings to notify you if Logical Disk: % Free Space, or Memory: Available Bytes falls below 20%, or if System: Processor Queue Length exceeds 3.
Save these counters in a workspace settings file, like Overview.pmw, so you can start them easily. They will provide a broad view of the performance of your computer.