Abstract: Techniques have been provided for storing data in a de-duplicated manner on a data storage system in a manner that allows for real-time reference to an index that is too large to fit within memory. This may be accomplished by segmenting the index into smaller segments, stored on disk. Only a subset of the segments may be loaded into memory at a given time. A predictive filter is stored in memory for each segment, allowing a de-duplication driver to quickly predict whether any given new block is likely to be indexed by each segment. Since identical blocks are often stored in long identical sequences (e.g., upon copying a disk image to a disk for a virtual machine), once a segment stored on disk is referenced many times in a short period, it is loaded into memory to allow the remainder of the long sequence to be de-duplicated.