Optical Burst Switching (OBS) was proposed ten years ago as an alternative switching paradigm in order to overcome some of the drawbacks of Optical Circuit Switching (OCS). While OBS is no more necessarily perceived as a competitor to OCS, but more of a more adapted switching for networks with bursty and highly dynamic traffic, there is still a debate around OBS, i.e., how far an OBS network can go in terms of throughput with no or limited burst losses. This thesis attempts to answer this question by investigating how to devise an upper bound on the throughput of an OBS network, assuming no recourse to electrical buffering is made at any intermediate node. We investigate both the burst scheduling and routing issues, with a larger focus on routing in three directions: (i) exploration of weighted k -shortest paths, (ii) revisiting load balancing, (iii) examining tree decomposition. Simulations have been conducted to compare and evaluate each of the new ideas with adapted (with respect to throughput upper bounding) previously proposed routing algorithms on different network and traffic instances. A comparison of the best upper bound with lower bounds obtained under various assumptions is presented