The discrete dipole approximation (DDA) is a variant of volume-integral equation method, which speed largely depends on the iterative solution of system of linear equations. I systematically studied the performance of this solution, varying the particle refractive index, DDA formulation (including non-standard ones), and several Krylov-subspace iterative solvers. For that I used publicly available ADDA code, so the conclusions can be directly employed by the practitioners of the DDA. Apart from the expected strong dependence on the refractive index, the number of iterations significantly differs between the DDA formulations, especially for purely real refractive indices. For small particles the number of iterations versus refractive index can be estimated by a simple relation, previously derived from analysis of the spectrum of the interaction matrix.