This paper considers two possible approaches, which enable multiple pairs of users to exchange information via multiple multiantenna relays within one time-slot to save communication bandwidth in low-latency communications. The first approach is to deploy full-duplexes for both users and relays to make their simultaneous signal transmission and reception possible. In the second approach, the users use a fraction of a time slot to send their information to the relays and the relays use the remaining complementary fraction of the time slot to send the beamformed signals to the users. The inherent loop self-interference in the duplexes and inter-full-duplexing-user interference in the first approach are absent in the second approach. Under both these approaches, the joint design of the users' power allocation and relays' beamformers to either optimize the users' exchange of information or maximize the energy-efficiency subject to user quality-of-service (QoS) constraints in terms of minimal rate thresholds leads to complex nonconvex optimization problems. Path-following algorithms are developed for their computational solutions. Numerical examples show the advantages of the second approach over the first approach.