Most administrators will agree that ordinary benchmark applications don't always reflect real-world application usage. And if you're running out-of-the-ordinary software, the benchmark results are pretty much useless anyway.
However, it's still important to find an application that monitors your organization's systems and reports the usage of the most important resources (e.g., CPU time and memory). Windows NT comes with an application that does all of this: Performance Monitor.
Let's look at how you can use Performance Monitor to establish a baseline and monitor the counters. You can find Performance Monitor by going to Start | Programs | Administrative Tools | Performance Monitor.
First, establish the baseline. The baseline represents the values of the counters for monitored resources during normal operation. For example, you should measure the CPU load when setting up your servers. Let's say the CPU load is 10 percent at the beginning. This is your baseline now.
Next, watch the counters as they change over time. If an increasing number of users use your server, the CPU load will increase—at first to 20 percent, then 50 percent, and so on.
By monitoring the counters, you can predict future upgrades. For example, if the CPU usage increases from 10 percent to 70 percent in one month, you can start planning the upgrade. You definitely don't want to start thinking about upgrading when your system is constantly at 99 percent.
If you want to correctly identify the bottlenecks you may encounter, you'll need to know which counters to monitor. Here are some of the counters that offer the most information about a system.
- Processor\%Processor Time: This counter represents the CPU usage. If this value is consistently more than 85, consider upgrading to a faster CPU. However, this counter can be very misleading. Sometimes the CPU isn’t too slow at all. For instance, if your hard drives work in PIO mode, the CPU usage will be very high during transfers. Using DMA mode will sharply decrease the CPU load and increase the transfer rate at the same time.
- System\Processor Queue Length: This represents the number of ready threads waiting in the CPU queue. A number larger than two usually indicates processor congestion.
- LogicalDisk\Avg. Disk Queue Length and PhysicalDisk\Avg. Disk Queue Length: These two counters display the average number of all requests (read and write) that are queued and waiting to be services for the selected logical partition and disk. If these values are greater than two over a long period of time, this might indicate a disk bottleneck.
- Memory\Pages/sec: This counter displays the number of pages per second. Paging occurs when the system can't find a requested page in the memory.
- Memory\Available Bytes: This counter displays the amount of physical memory available to processes running on the system. Add additional memory if this counter is constantly low.