Resumen
The article presents an experience of reengineering and tuning large client/server architecture (CSA). CSA is introduced to overcome issues of distributed computing such as remote data access, splitting database and network latency. For large scale systems, workload model is typically stated in terms of probability distributions where a user of a particular type generates a probabilistic number of queries per unit time. The performance model is used to analyze the current system performance and predict future performance under varying workload and architecture changes. It takes into account the physical hardware, connectivity, architecture, software and workload model. Cost is frequently modeled as a constraint, or as part of an objective function. Observability of the service demands is often an issue due to the extreme complexity of large-scale distributed systems. Load testing is considered a good approach for both performance and scalability analysis as well as collecting data as input for performance models.