Clouds are increasingly being used for running dataintensive scientific applications. However, science applications need to contend with the I/O and network performance characteristics of cloud environments. Additionally, managing data effectively and efficiently over these cloud resources is challenging due to the myriad storage choices with different performance-cost trade-offs, complex application choices, complexity associated with elasticity and failure rates. In this paper, we evaluate various aspects of data management strategies in cloud environments. Our evaluation is performed in the context of two frameworks - Hadoop and FRIEDA and conducted on four cloud testbeds - FutureGrid, ExoGeni, Grid5000, Amazon. Our experiments highlight the different performance implications of storage, file system, provis