Techniques for configuring multi-tier data storage systems provide a model of skewed workload distributions in such systems and configure them by applying a received input skew value representing a prediction of system skew and the model. The technique applies the input skew value to generate a desired proportion of total system IOPS to be handled by each storage tier in the system and applies the model to convert the IOPS percentage for each storage tier into a corresponding capacity percentage. The technique then generates actual capacity for each tier by applying the percent capacity for each tier to a design target for total storage capacity in the system. Advantageously, these techniques generate storage tier configurations that are likely to be neither overdesigned nor underdesigned, and are thus likely to meet both performance requirements and cost targets.