Efficient tamper-evident data structures for untrusted servers.

摘要

Many real-world applications run on untrusted servers or are run on servers that are subject to strong insider attacks. Although we cannot prevent an untrusted server from modifying or deleting data, with tamper-evident data structures, we can discover when this has occurred. If an untrusted server knows that a particular reply will not be checked for correctness, it is free to lie. Auditing for correctness is thus a frequent but overlooked operation. In my thesis, I present and evaluate new efficient data structures for tamper-evident logging and tamper-evident storage of changing data on untrusted servers, focussing on the costs of the entire system.;The first data structure is a new tamper-evident log design. I propose new semantics of tamper-evident logs in terms of the auditing process, required to detect misbehavior. To accomplish efficient auditing, I describe and benchmark a new tree-based data structure that can generate such proofs with logarithmic size and space, significantly improving over previous linear constructions while also offering a flexible query mechanism with authenticated results.;The remaining data structures are designs for a persistent authenticated dictionary (PAD) that allows users to send lookup requests to an untrusted server and get authenticated answers, signed by a trusted author, for both the current and historical versions of the dataset. Improving on prior constructions that require logarithmic storage and time, I present new classes of efficient PAD algorithms offering constant-sized authenticated answers or constant storage per update. I implement 21 different versions of PAD algorithms and perform a comprehensive evaluation using contemporary cloud-computing prices for computing and bandwidth to determine the most monetarily cost-effective designs.