Adaptive Sharding and Fault Tolerance in Distributed Systems Bhavana Mehta

Format:

Book

Thesis/Dissertation

Author/Creator:

Mehta, Bhavana, author.

Contributor:

University of Pennsylvania. Computer and Information Science., degree granting institution.

Language:

English

Subjects (All):

0489.

0537.

0800.

0984.

Local Subjects:

0489.

0537.

0800.

0984.

Physical Description:

1 electronic resource (141 pages)

Contained In:

Dissertations Abstracts International 87-07B

Place of Publication:

Ann Arbor : ProQuest Dissertations and Theses, 2025

Language Note:

English

Summary:

Achieving scalability and reliability in distributed database systems requires effective data fragmentation and replication. In Byzantine environments, where nodes may behave adversarially, traditional partitioning schemes no longer hold. This dissertation introduces an adaptive sharding framework that combines dynamic data partitioning with Byzantine fault tolerance to operate securely in untrusted environments. We use a progressive architectural approach. First, we establish a baseline with an optimized centralized architecture for environments with limited trust assumptions. Building on this, we develop our primary contribution: a decentralized architecture that eliminates single points of failure in adversarial settings where no trusted components exist. Both architectures perform adaptive resharding: the centralized baseline uses a global hypergraph partitioner, while the decentralized design uses our transaction-aware key-affinity partitioning algorithm, which quantifies relationships between keys and shards to make resharding decisions based on workload patterns. The system achieves consensus through Practical Byzantine Fault Tolerance (PBFT) for both intra-shard and cross-shard transactions. Our evaluation shows that both architectures outperform static, non-resharding baselines. The decentralized design achieves up to 2.5× higher throughput than a non-resharding baseline and remains effective even as the fraction of cross-shard transactions increases. By redistributing data adaptively, both designs mitigate the impact of adversarial behavior while maintaining fault tolerance. Testing against Byzantine attacks shows 68% maintenance of fault-free throughput with 25% Byzantine nodes. Our contributions include Byzantine-aware adaptive sharding algorithms, secure cross-shard transaction protocols, and evaluations suitable for applications in decentralized finance, cloud computing, and IoT environments.

Notes:

Advisors: Loo, Boon Thau; Amiri, Mohammad Javad Committee members: Marcus, Ryan; Smith, Jonathan M.; Liu, Vincent

Source: Dissertations Abstracts International, Volume: 87-07, Section: B.

Ph.D. University of Pennsylvania 2025

Vendor supplied data

Local Notes:

School code: 0175

ISBN:

9798276001913

Access Restriction:

Restricted for use by site license