Dimensioning On-Demand Vehicle Sharing Systems

We consider the problem of optimal fleet sizing in a vehicle sharing system. Vehicles are available for short-term rental and accessible from multiple locations. A rented at one location can be returned to any other location. The size must account not only nominal load randomness demand duration but also number vehicles that each because roaming (vehicles returning same which they were picked up). model dynamics system using closed queueing network obtain explicit form lower upper bounds on (the minimum needed meet target service level). Specifically, we show starting with pair bounds, always another gaps between independent bounded by function depends prescribed level. generated asymptotically exact under several regimes. use features construct simple approximation within is asymptotic regimes considered. Extensive numerical experiments approximate values nearly indistinguishable wide range parameter values. highly interpretable buffer capacity expressed terms three interpreted as follows: (1) standard protection against times, (2) roaming, (3) correction term. Our analysis reveals important differences systems (where servers return queue upon completion) where servers, completion, randomly join queues additional substantial even vanishingly small demand. This paper was accepted Baris Ata, stochastic models simulation.