Scalable High-Throughput Systems for Practical Access Privacy

The massive amounts of data stored and processed in shared (often
public) online spaces raises essential security and privacy concerns.
Data encryption is usually the first step towards ensuring some form of
privacy. However, data access patterns can reveal a wealth of
information about encrypted data. This is because applications typically
have data-dependent memory access patterns i.e., the order in which
memory locations are accessed is determined by the data.

To mitigate this, oblivious RAMs (ORAM) have been proposed as a
solution. Yet, ORAMs are not considered viable for practical
deployments due to their prohibitive overheads. These overheads are
often exacerbated by design choices that overlook important performance
metrics. This talk shows that by leveraging
application-specific security-performance trade-offs, we can design
significantly more efficient access-privacy-preserving systems. To
demonstrate this, I will discuss two ORAM constructions
which specifically address two important performance challenges: i)
enabling parallel oblivious multi-client access for shared data, and ii)
enabling oblivious locality-preserving access to significantly speedup
locality-optimized applications e.g., file systems.
As a key take-away, this talk highlights the importance of
re-thinking foundations and optimizing solutions for metrics that are critical for
performance on real hardware (e.g., locality of access) for building
efficient and scalable secure systems.


Bio:
Anrin Chakraborti is a PhD candidate in Computer Science at Stony Brook University,
where he is a member of the National Security Institute (NSI) and the Network Security and Applied Cryptography Lab.
His research involves building efficient and provably secure systems to address critical data privacy problems.