It is now well-known that for a mobile ad hoc network (MANET), interference alignment (IA) beamforming can greatly improve the network sum throughput. In a real implementation, an IA beamforming algorithm should be combined with other techniques, such as link selection and power control, while the existing IA algorithms ignore these practical aspect. In this paper, we present a new IA beamforming algorithm for maximizing the network sum throughput by using a sequential convex programming (SCP). We show that in maximizing the sum throughput of a network consisting of $N$ contending links, this distributed algorithm can automatically select $K$ ($Kle N$) out of them, and the $K$ selected links should use the maximum transmit power.