ZEN AND THE ART OF PERFORMANCE MONITORING 373 . Monitor activity during typical periods of activity. The system will definitely look underused if you monitor these same resources at midnight. However, I recommend monitoring activity at midnight just to see how much of the major resources are actually being used when there is little to no activity on the server and to create an idle baseline. . Don t sweat the peaks. Look for sustained activity throughout the life of your monitoring session. Spikes in activity such as CPU and disk usage are standard and nothing to be concerned about. It is normal for a CPU to spike to 100 percent occasionally. . Establish a baseline for your system. Monitor your system when there are no active users to see how the system components such as memory, disk access, processor utilization, and network usage behave when the system is in an idle state. Then monitor your system as its load increases. Make sure to save the log files so you can look at historical data and compare it with your projected system growth and system resource requirements. . Maintain performance information over time to show growth impacts on the system and application availability. Performance Checks: Quick and Dirty If I have a performance problem on any server, the Performance console is not the first tool I fire up. One of the coolest additions to Windows NT 4 was the addition of the Windows Task Manager. You can quickly glean lot of useful information from the Task Manager, but be warned that this information is at-a-glance information and has no bearing on performance history beyond what you see on the screen. My favorite starting place is the Performance property page. Figure 9.1 shows the Performance property page from a typical Exchange 2003 server. This server has 2GB of RAM; you can discern that information from the Physical Memory s Total counter. Of this RAM, approximately 927MB of it is available to other processes, and approximately 1GB of this RAM is in use by the system cache. Monitoring Is Not a One-Time Occurrence When you begin performance monitoring of any kind, you are concerned about averages, not spikes in activity. How do you get this average? You watch your systems over long periods of time. The performance data you gather from a single morning of monitoring may not be representative of your server s true behavior. The performance data you gather over a week may not even give you a true indication of the normal behavior of server performance. Only after taking samples for weeks at a time will you be able to get a better idea of average usage. Taking a single look at an Exchange server and deciding from that single look that you have a certain type of bottleneck might yield accurate results. But then again, it may yield an incorrect solution based on incomplete information. You may spend thousands of dollars to fix a bottleneck that did not really exist in the first place. In the meantime, the real bottleneck goes on unresolved. When you begin performance monitoring, have a long-term goal in mind. Learn to use the analysis tools in a product like Microsoft Excel, or purchase a monitoring software package such as Microsoft Operations Manager ( www.microsoft.com/mom ) or NetIQ ( www.netiq.com ) that can help you to collect, store, and analyze Exchange performance data.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Low Cost Web Hosting services