This dissertation considers the problem of scaling ad hoc wireless networks now being applied to urban mesh and sensor networks scenarios. Ad hoc networks involve multi-hop communication which has inherent scaling problems in that throughput per node drops as the square root of the number of nodes in the network. We investigate mechanisms for improving performance and scalability of multi-hop wireless networks, with focus on system architecture and routing protocol aspects.First we propose a generalized multi-tier hierarchical hybrid network with three tiers of radio nodes: low-power end-user mobile nodes (MN) at the lowest tier, higher power radio forwarding nodes (FN) that support multi-hop routing at intermediate level, and wired access points (AP) at the highest level. We present an analytical model for the capacity of the proposed network and identify conditions on transmission range and node density for scalability to be maintained. From the derived upper and lower bounds, it is shown that the low-tier capacity increases linearly with the number of FN's, and that the high-tier capacity grows linearly with the number of AP's in the scaling region.The analytically obtained capacity results are validated with detailed system simulations for dense network scenarios. The simulation study also examines the allocation of separate channels to avoid the increased protocol overhead which arises in the single channel case. A heuristic distributed channel assignment algorithm is proposed to achieve conflict-free transmissions in the network.Next, we investigate cross-layer adaptive routing as another type of scaling mechanism. An adaptive routing framework, which allows introduction of adjustable parameters and programmable routing modules, is described. The proposed framework can support various cross-layer mechanisms including those based on integrated routing metrics that incorporate PHY and MAC information.We investigate a PHY/MAC aware routing metric (PARMA) which incorporates physical layer link speed and MAC congestion. Design and implementation of PARMA are outlined, and simulation results for typical multi-rate 802.11 ad hoc network scenarios show that PARMA helps improve throughput and decrease congestion by selecting paths with high bit-rate links while avoiding MAC congestion areas.