We examine the primary challenges for building a practical and competitive holographic random access memory (HRAM) system, specifically for size, speed, and cost. We show that a fast HRAM system can be implemented with a compact architecture by incorporating conjugate readout, a pixel-matched sensor array, and a linear array of laser diodes. It provides faster random access time than hard disk (100 microseconds or less) and similar bandwidth as silicon storage with lower cost. Preliminary experimental results support the feasibility of this architecture. Our analysis shows that in order for the HRAM to become competitive, the principal tasks will be to reduce spatial light modulator (SLM) and detector pixel sizes to 1 mum, increase the output power of compact visible-wavelength lasers to several hundred milliwatts, and develop ways to raise the sensitivity of holographic media to the order of 1 cm/J.