Packet-scheduling is a particular challenge in wireless networks due to interference from nearby transmissions. A <i>distance-2 interference model</i> serves as a useful abstraction here, and we study packet routing and scheduling under this model. The main focus of our work is the development of fully-distributed (decentralized) protocols. We present polylogarithmic/constant factor approximation algorithms for various families of disk graphs (which capture the geometric nature of wireless-signal propagation), as well as near-optimal approximation algorithms for general graphs. The packet-scheduling work by L eighton, Maggs and Rao (<i>Combinatorica</i>, 1994) and a basic distributed coloring procedure, originally due to Luby (<i>J. Computer and System Sciences</i>, 1993), underlie many of our algorithms. Experimental work of Finocchi, Panconesi, and Silvestri (SODA 2002) showed that a natural modification of Luby's algorithm leads to improved performance, and a rigorous explanation of this was left as an open question; we prove that the modified algorithm is provably better in the worst-case. Finally, using simulations, we study the impact of the routing strategy and the choice of parameters on the performance of our distributed algorithm for unit disk graphs.