Queuing Networks for Designing Shared Services

Shared services have been widely implemented in many government and business organizations (Janssen & Joha, 2006; Leavell, 2006; Rolia, Cherkasova, Arlitt, & Machiraju, 2006; Ulbrich, 2006; Wang & Wang, 2007; Williams, 2006). Government and large business organizations often have many operational units. Each operational unit provides and supports particular business functions to the organization. For shared services, common business functions are consolidated and shared by different operational units within the organization in order to reduce operational cost and to increase information and knowledge sharing. Shared services generally require significant transformation and optimization of business processes. Cost saving and knowledge sharing within the organization are the strategic opportunities of shared services (Davenport, Harris, & Cantrell, 2004; Davenport & Short, 1990). One of the important aspects of successful shared service design is to maintain specified service levels for each operational unit that uses the shared service. Different operational units have different requirements about the completion time of the tasks (Wang, 2007). It is important and necessary to ensure all these requirements are fulfilled when designing shared services. Multi-class product-form queuing networks have been widely used as an analytic tool for predicting the completion time of tasks in a system (Bolch, Greiner, Meer, & Trivedi, 2006). They can facilitate the successful design of shared services by checking whether all requirements of different operational units are fulfilled (Wang, 2007). Checking whether all requirements of different operational units are fulfilled means to solve the corresponding multi-class product-form queuing network model of the shared service. However, as the complexity of shared services increases, the computational time for solving a multi-class product-form queuing network model becomes prohibitive large. Various approximate Mean Value Analysis algorithms have been proposed to solve multi-class product-form queuing network models with much reduced computational time and a reasonably accuracy (Wang & Sevcik, 2000; Wang, Sevcik, Serazzi, & Wang, 2008). In practice, these approximate algorithms can be used together with multi-class product-form queuing network models as an analytic tool for facilitating the successful design of shared services. Hai Wang Saint Mary’s University, Canada