Yggdrasil: Privacy-Aware Dual Deduplication in Multi Client Settings

Bonsai: A Generalized Look at Dual Deduplication

“…For each received chunk of file, if the chunk is already present in the set of bases the CSP deduplicates it by storing a file identifier and a pointer representing the base. Otherwise, the CSP looks for a base similar to the given chunk, i.e., a chunk in the set of bases that has a short distance to the received chunk, e.g., according to the Hamming distance [9] or the Swap distance [10]. If such a base is found, the CSP deduplicates this record in a generalized way by storing a file identifier, a pointer representing the closest base, and a string containing the difference between the base and the chunk.…”

“…Recently, the work by Sehat et al [10] bypasses the programmable random oracle challenge and introduces a unique privacy-aware deduplication mechanism for multi-client environments called dual deduplication. In dual deduplication, clients perform information-theoretic transformations on their data before uploading it to the CSP.…”

“…These transformations aim at puncturing the data to simultaneously increase deduplication capabilities on the CSP, and creating uncertainty for the CSP about the possible pre-images of the uploaded chunk. From a high level perspective, to the eye of a CSP or an attacker, the dual deduplication system proposed in [10] mimics a deletion channel from information theory [14]. In particular it is known that retrieving the original data after transmission through the deletion channel is a hard problem.…”

“…While these properties are undesirable to establish a reliable communication between two parties, dual deduplication uses them as key features for proving the privacy of the system. In this work, we build on the dual deduplication idea introduced in [10], and mitigate the three major drawbacks of the original construction. First, clients in [10] need to store a significant portion of their data locally.…”

“…In this work, we build on the dual deduplication idea introduced in [10], and mitigate the three major drawbacks of the original construction. First, clients in [10] need to store a significant portion of their data locally. This is contrary to the intent of outsourcing data to a CSP and minimizing local storage.…”