Coding and data allocation schemes for reliable, secure, and resource-efficient distributed systems

Abstract

We are living in an era of data deluge where more than 2.5 Exabytes of data are created every single day. In order to deal with the increasing demand for processing big data, centralized systems have been progressively replaced by distributed systems where multiple nodes located on different machines collectively act as a single coherent system. However, distributed systems suffer from reliability and security issues due to inevitable node failure events and adversarial attacks. This thesis focuses on constructing reliable, secure, and resource-efficient distributed systems based on coding and information theory. In the first part, we obtain the secrecy capacity, the maximum data size that can be stored with perfect secrecy, in clustered distributed storage system and provide explicit coding schemes to securely store the data against eavesdroppers. In the second part, we suggest a general framework for the coded Practical Byzantine Fault Tolerance consensus algorithm for enabling resource-efficient agreement among distributed nodes under Byzantine attacks. In the last part, we propose communication-computation efficient secure aggregation which considerably reduces the amount of communication/computational resources required for private federated learning.