SPARC-Stream: A Sketch-Based Privacy-AwareResilient Compression Framework for High-Throughput Graph Stream Analytics

Abstract

Modern data-intensive applications generate massive graph-structured data streams from social networks, financial systems, sensor infrastructures, and online services. Extracting meaningful insights such as heavy hitters, dense communities, and neighborhood relationships from these streams requires algorithms that are memory-efficient, privacy-aware, and capable of operating under dynamic conditions involving insertions and deletions. Existing approaches address individual challenges such as heavy hitter detection, dense subgraph discovery, or compression independently, but fail to provide a unified framework capable of supporting multiple analytical tasks while ensuring persistence efficiency and privacy guarantees. This paper proposes SPARC-Stream (Sketch-based Privacy-Aware Resilient Compression Stream Framework), a novel architecture designed for scalable graph stream analytics. The proposed technique integrates four key components: (i) a Universal Graph Sketch that simultaneously estimates heavy hitters, frequency moments, and neighborhood structures; (ii) a Compressed Time-Series Edge Encoding module for efficient floating-point data representation; (iii) a Persistent Epoch Buffer inspired by NVM- based buffered persistence; and (iv) a Differentially Private Neighborhood Estimator. Together, these components enable accurate and efficient online analytics over high-volume graph streams. Extensive experimental evaluation demonstrates that SPARCStream improves analytical throughput by up to 2.7x while reducing memory consumption by 45% compared to state-of-the-art techniques.