SORS: a scalable online ridesharing system

In this paper, we design and evaluate SORS- a scalable online ridesharing system, where drivers and passengers send their requests for a ride in advance, possibly on a short notice. SORS is modular and consists of two main, loosely coupled, components: the Constraint Satisfier and the Matching Module. The Constraint Satisfier takes as input information about the desired trajectories and spatio-temporal constraints of drivers and passengers and it returns a list of feasible (driver, passenger) pairs. We use a road networks data structure, optimized for the specific spatio-temporal queries in the context of ridesharing, and we show that our Constraint Satisfier has a 4.65x more scalable query time than a general-purpose database. We represent the feasible pairs of drivers and passengers as a weighted bipartite graph with edge weight being the length of the shared trip of the pair, which captures the revenue in real-world ridesharing systems, such as Lyft Carpool. The Matching Module then takes as input this weighted bipartite graph and returns the maximum weighted matching (MWM), using an algorithm that solves the problem online and efficiently, by incrementally updating the matching solution in real-time. We show that our algorithm achieves 51% larger weight (i.e., total revenue) compared to greedy heuristics used by many real systems today. We also evaluate the SORS system as a whole, using mobile datasets to extract driver trajectories and passenger locations in urban environments. We show that SORS can provide a ridesharing recommendation to individual users within a sub-second query response time, even at high workloads.