This paper focuses on studying orthogonal and non-orthogonal multiple access in intelligent reflecting surface (IRS)-aided systems. Unlike most prior works assuming continuous phase shifts, we employ the practical setup where only a finite number of discrete phase shifts are available. To maximize the sum rate, active beamforming and discrete reflection need to be jointly optimized. We therefore propose an alternative optimization method to get the optimal continuous phase shifts iteratively, and then quantize each phase shift to its nearest discrete value. The sum-rate maximization of different schemes is theoretically analyzed and numerically evaluated with different numbers of phase-control bits.