We are concerned with the throughput of a full-duplex (FD) MIMO network. Unlike conventional half-duplex (HD) MIMO, two wireless devices of a bidirectional FD-MIMO link have freedom of selecting which antennas/RF-chains to transmit or receive before tuning their radiation patterns to maximize the link's throughput. The freedom in configuring the function of available RF-chains, resulting in various FD-MIMO transmission modes, is referred to as FD-MIMO freedom that is shown to significantly improve the spectral efficiency of a given link. For a given RF-chain/antenna selection of a set of FD-MIMO links, we end up with a non-convex throughput maximization problem of a heterogeneous MIMO network. We design both centralized (using the augmented Lagrange function) and distributed algorithm (using a hierarchical game and pricing) to solve the problem for its locally optimal solutions. Comparing the achieved throughput of the FD-MIMO network, averaged over all obtained locally optimal solutions, with that when FD-MIMO nodes choose to operate in an HD mode, we find the HD mode surprisingly outperforms the FD mode. This trend is also observed when exploring all possible communication modes of a small size FD-MIMO network.