From a queuing theory perspective, when the workload is better balanced, the arrival rate is reduced. Figure 9-20 graphically shows when the arrival rate is decreased, we moved from point A to point B. When the arrival rate is decreased, system requirements decrease, resulting in a utilization decrease, as well as a response time decrease. Unlike with the tune and buy options, there is no response-time curve shift. What has shifted is the workload; that is, the system has traveled along the response-time curve. This is usually difficult to initially understand. But consider that when the workload has decreased, there is no change in the service time, as it takes a transaction processor just as long to process a single transaction as before. Therefore, the response-time curve does not shift down. The response-time curve does not shift to the right because no additional capacity has been implemented. What changed is the arrival rate, so we simply move along the existing response-time curve as the response time decreases. As we move to the left, while service time remains the same, the queue time will decrease, resulting in an improved response time.

Figure 9-20. Shown is the effect of workload balancing on response time. During peak processing time, the response time is at point A. By shifting some of the workload to another time, the arrival rate is reduced from 24 trx/ms to 18 trx/ms, resulting in a significant response time reduction. Notice the response-time curve does not shift, but rather the system activity has shifted.

Now, we finally answer the question posed at the beginning of the chapter: What kind of performance improvement can we expect? I'll start with some direction and words of caution. Then I will move directly into a series of exercises to demonstrate step by step how you can creatively apply all the material presented in this book to anticipate the impact of a solution and improve the situation.