Dynamic balancing in finite processor sharing queues with guard bandwidth policy, multiclass retrial users and signals

We analyze a novel multiclass two-station retrial system with processor sharing to model the access of elastic traffic in modern mobile networks managed by two service providers. We focus on a small region containing a single access point per provider, which admits a limited number of users, say the handover users, the subscribers and flex users. Fresh subscribers join their provider, while fresh flex users connect with the provider, which offers the largest data rate. Blocked handover users are lost, whereas blocked fresh subscribers retry independently to connect with their provider after a random amount of time. Blocked flex users retry also to connect with the provider that offers the largest data rate at their retrial instant. In each access point the service rate depends on the type and the number of users, and on the state of a random environment, whereas a guard bandwidth policy to prioritize the handover users is employed. Both access points cooperate by allowing the seamless transition of an ongoing file transfer to the “fastest” provider. We calculate the joint stationary distribution of the number of users of each type in each access point, and study sojourn times of a connected user, conditional both on the state upon arrival and on the work brought to the system. Waiting time analysis for the blocked retrial users of either type is also investigated. Numerical results are obtained and illustrate the system's performance.