The following list describes objects and counters that may be useful for evaluating demand and performance on your Windows NT computer. For more information see your Windows NT documentation.
Memory: Pages/sec
In order to understand memory load on a Windows NT Server, it is necessary to understand paging, a technique for implementing virtual memory. Paging is the swapping of blocks (pages) of program instructions or data back and forth between memory and disk, and is carried out as needed by the virtual memory manager in Windows NT.
Pages/sec is the number of pages read from the disk or written to the disk because they were not in memory when needed (that is, the number of "page faults" that required disk access.) The counter includes paging traffic generated when the cache accesses file data for applications.
This is the primary counter for determining whether your server is paging excessively (or "thrashing".) As this counter goes up, server responsiveness slows because of the time required for disk access (reading or writing). A server dedicated to communications should be equipped with enough physical memory so that little paging is required.
The highest acceptable value for Pages/sec varies from system to system. One way to judge whether system load is causing too much paging is to observe whether processor activity (System: %Total Processor Time) drops significantly as paging increases. This indicates that the system is occupied with swapping pages rather than with actually processing instructions.
The primary ways to correct excessive paging is to add more physical memory to the server or to consider ways to decrease the demand on the server. Demand can be decreased by narrowing the variety of tasks that a server must perform or by decreasing the number of users accessing a server. For example, an overloaded multipurpose server (with file, print, and SNA Server demands) could be dedicated to SNA Server traffic only. Or, as another example, user loads placed on one server could be divided between two servers (load balancing). As mentioned, in any case of memory overload, adding physical memory may provide the needed performance increase.
A secondary way to decrease the impact of paging is to upgrade the disk system to be more efficient. This includes installing a faster disk, installing a second disk, using RAID striping, or similar upgrades. This upgrading does not decrease paging (Pages/sec), but speeds up the paging process itself. For example, replacing a slow IDE disk with a faster SCSI disk may make a given paging rate (perhaps 20 – 40 pages/sec) acceptable.
System: %Total Processor Time
Processor: %Processor Time
System: %Total Processor Time is the percentage of elapsed time during which the system processors are busy. It can be viewed as the fraction of total processor time spent doing useful work. Values of 60 – 80 percent during typical loads are good values, because they allow some reserve for peak loads. However, when the processor stays at 100 percent for periods of time, this may indicate a processor bottleneck. On a multiprocessor system, you can view Processor: %Processor Time for each processor, to see how the load is distributed among processors.
One useful way to view Total Processor Time values is in Chart view along with counters indicating increases and decreases in user load. For user load, such counters include logical unit sessions: Throughput Bytes/Sec, and SNA Adapter adaptername: Throughput Frames/Sec. These two counters are available only when there is SNA activity. For example, you might notice that during a period of peak demand for logical unit sessions, Total Processor Time reaches 100 percent and stays there. This could indicate that the SNA Server is reaching peak capacity and that any future additional demand might require additional processors or additional servers.
It may also be helpful to view System: %Total Processor Time along with any other counters related to the server's major functions. For example, for an SNA Server that is also a file server, the Server: Server Sessions counter may be helpful. Other counters that may help you analyze the sources of processor activity are Process: %Processor Time for processes you think are relevant, as well as System: Total Interrupts/sec. Also, if your clients are using NetWare, consider looking at NWLink SPX: Connections Open. Similarly, if your clients use TCP/IP, consider looking at TCP: Connections Established.
System: Total Interrupts/sec
Processor: Interrupts/sec
System: Total Interrupts/sec is the rate at which the computer is receiving and servicing device interrupts. (Device interrupts are the signals a device sends to a processor to indicate that a task is complete or that the device requires attention.) Some devices that may generate interrupts are adapters, network adapters, the system timer (clock), and the mouse. System: Total Interrupts/sec provides an indication of how busy these devices are on a computer-wide basis.
Similarly, for each processor, Processor: Interrupts/sec is the rate at which the processor is receiving device interrupts.
Normal thread execution is suspended during interrupts. An interrupt may cause the processor to switch to another, higher-priority thread. Clock interrupts are periodic and frequent (on the order of 100/sec); they create a background of interrupt activity.
These counters can help indicate the general demand on a server, and may be useful when combined with processor and memory data such as System: %Total Processor Time and Memory: Pages/sec.