Advanced reservation network architectures for scientific applications: An algorithmic perspective

As a result of the current limitations of TCP/IP architectures in supporting high-throughput scientific and other grid applications, significant efforts have recently been made to complement them with new architectures enabling advanced reservation of dedicated resources. In this chapter, we survey on-going work (including our own) in this emerging research area. We characterize the capacity bounds of advanced reservation architectures and describe two new principles, called path grading and path switching, for the design of such architectures. We present a polynomial-time algorithmic framework, called Graded Channel Reservation (GCR), implementing these principles. Our findings are supported by numerical results illustrating the utility of path switching and path grading and comparing the performance of GCR and its variants to the capacity bounds. The chapter is concluded by giving some insights to potential future grounds for the presented GCR algorithms.