Method and apparatus to recover from interrupted data streams in a deduplication system

Detection and proper deduplication of a re-started data stream in a segmentation analysis-based deduplication system are provided by retaining information about a previous data stream and using that information when performing segmentation of the re-started data stream. Information such as a segment size associated with a last data object received in the previous data stream and a record of how much data was present in the last segment associated with the previous data stream is retained. The retained segment size information is used to set a first data object segment size of the re-started data stream, and the size of last segment information is used to determine how much information should be put in the first segment associated with the re-started data stream in order to maintain proper alignment of the remainder of the segments for the first data object in the re-started data stream for deduplication.