Attempts to improve database and operating system performance are best done using careful observation, not speculation, in order to determine where the system bottlenecks are. You need to start that before you have a performance problem, to record baseline information. Most production database servers should consider basic monitoring and trending setup a requirement of their early deployment.
However, it is helpful to know the low-level tools too because the longer-term views provided by most monitoring and trending tools will miss brief problems. With today's breed of database applications, even a pause lasting a few seconds could be a major response time failure, and it's one that you wouldn't even be able to see in data collected on a minute scale. Both short-term and long-term data collection has considerable value, and knowing when to switch the detail...
