Summary performance

The best way to measure performance of a machine happens if the administrator is able to compare the performance indicators on various moments. This is possible if a baseline is created for various parameters that show the resource usage of a system. 

You can use the system monitor to measure system performance. This system monitor displays the performance parameters for a system and is part of the performance console. (perfmon.msc) It can be start from the Administrative tools program group (Windows 2000 Prof.) and can show performance information for a local or remote computer in figures, charts, histograms or reports. The console is also part of the computer management console.

The type of data delivered by the system monitor is based on objects, counters and instances. An object is a part of the computer for which data is collected, e.g. the hard disk. A counter is one of the performance values collected for that object, e.g the percentage of disk time or the available bytes. The instance tells for which of the objects the data is collected if there are more than one of these objects in the system. E.g. disk 1 or disk 2.

When looking for performance bottlenecks, the performance objects can be split up in four areas :

Processor

The performance counter %Processor Time gives the best view on the processor usage. The usage will climb to 100 percent when applications starts or on peak loads. If the average of the processor load is higher than 80 %, a process upgrade or the adding of a new processor is recommendable. Another parameter that might show processor issues is Processor queue length. If it's bigger than two there is a bottleneck.

Disk

Disk performances is measured for logical and physical disks. Physical disks are the disks within the system, logical disks are the volumes created within Windows 2000. Before you can collect logical disk performance figures, the measuring should be activated with the ‘diskperf’ command and the system should reboot. This command has the following options : 

Counters that can indicate a disk performance problem are %Disk Time (should not be higher than 90 %), Current disk queue length (should not be higher than the number of spindels + 2), disk reads per second and disk writes per second. A value higher than 0.3 seconds for Average Disk sec/Transfer may indicate a disk controller retrying the disk because of failures. To get a good impression of the value, usage the average of a longer measurement.

If disks become the bottleneck, performance can be increased by implementing (hardware) RAID, installing faster disks (on different controllers) or by defragging the disks.

Memory

Memory can be split in physical memory, the memory chips in the machine, and virtual memory. The virtual memory consists of the physical memory and the swap file. When there is not enough physical memory Windows swaps 4 KB blocks of memory (pages) to the swap file. This is called paging and paging causes a decrease of the system performance as all pages have to be written to disks and swapped back to memory when needed.

Indicators that can display a shortage of memory are Available Bytes and Pages per second. If the number of Available Bytes, the amount of memory for applications, drop below 4 MB or if the amount if pages per second is above 20, a memory problem seems to be there. To solve this, multiple page files on multiple disks can be created or the amount of memory can be increased. Of course, you can also decrease the number of applications on the system. 

Network

The best way to monitor potential network problems is to check the performance per OSI layer.  Object per layer are :

Performance logs and alerts

The performance log can be used to log performance data or events from (remote) systems. The collected data can be saved as tab- or comma-delimited files. The binary files are .blg files that are by default stored in the c:\PerfLogs directory. Performance logs are activated via the Performance snap in. The settings can be saved as .htm or report files (*.tsv). 

Counter logs

One kind of performance log is a counter log. These files contain performance information for parameters and intervals that are set by the administrator. The tabs contains the following information :

Trace logs

Trace logs are used to trigger events and are stored in sequential trace files or circular trace files with a .etl extension. Trace logs record data collected by operating system providers or non system providers and can be used by third party tools for system analyses.

Alerts

Alerts are created to start an action when a performance indicators exceeds a threshold. If this occurs a log entry in the application log can be made, a network message can be send, a performance data log can be started or a program can be activated.  

More information

Related white papers

Links

Last update : 17 July 2002