This paper investigates the sum-rate gains brought by power allocation strategies in multicell massive multiple-input- multiple-output systems, assuming time-division duplex transmission. For both uplink and downlink, we derive tractable expressions for the achievable rate with zero-forcing receivers and precoders, respectively. To avoid high-complexity joint optimization across the network, we propose a scheduling mechanism for power allocation, where, in a single time slot, only cells that do not interfere with each other adjust their transmit powers. Based on this, corresponding transmit power allocation strategies are derived, aimed at maximizing the sum rate per cell. These schemes are shown to bring considerable gains over equal power allocation for practical antenna configurations (e.g., up to a few hundred). However, with fixed number of users N, these gains diminish as M → ∞, and equal power allocation becomes optimal. A different conclusion is drawn for the case where both M and N grow large together, in which case improved rates are achieved as M grows with fixed M/N ratio, and the relative gains over the equal power allocation diminish as M/N grows. Moreover, we also provide applicable values of M/N under an acceptable power allocation gain threshold, which can be used to determine when the proposed power allocation schemes yield appreciable gains and when they do not. From the network point of view, the proposed scheduling approach can achieve almost the same performance as the joint power allocation after one scheduling round, with much reduced complexity.