Abstract: Whole slide images(WSIs) are very large (30-50GB each in uncompressed format), multiple resolution tissue images produced by digital slide scanners, and are widely used by pathology departments for diagnostic, educational and research purposes. Content-based Image Retrieval (CBIR) applications allow pathologists to perform a sub-region search on WSIs to automatically identify image patterns that are consistent with a given query patch containing cancerous tissue patterns. The results can then be used to draw comparisons among patient samples in order to make informed decisions regarding likely prognoses and most appropriate treatment regimens, leading to new discoveries in precision and preventive medicine. CBIR applications often require repeated, random or sequential access to WSIs, and most of the time the images are preprocessed into smaller tiles, as it is infeasible to bring the entire WSI into the memory of a computer node. In this study, we have designed and implemented a distributed deep memory hierarchy data staging system that leverages Solid-State Drives (SSDs) and provides an illusion of a very large memory space that can accommodate big WSI datasets and prevent subsequent accesses to the file system. An I/O intensive sequential CBIR workflow for searching cancerous patterns in prostate carcinoma datasets was parallelized and the I/O paths were altered to include the proposed memory system. Our results indicate that the parallel performance of the CBIR workflow improves and our deep memory hierarchy, staging framework produces negligible overheads for the application performance even when the number of staging servers and their memory sizes are limited.