The world is in the midst of an unprecedented growth of interconnected data, and graph processing systems are expected to play a vital role. Conventional graph algorithms designed for static graphs struggle to efficiently handle the continuous changes and updates that occur within these networks. As these networks grow in complexity, the need for algorithms capable of efficiently analyzing dynamic graph data is increasingly crucial. Our research aims to address the computational challenges posed by the need for real-time insights and scalable processing in dynamic and complex networks.

However, many dynamic algorithms are sequential, tailored towards web graphs, do not utilize reducibility, locality benefits of SCCs, overestimate affected vertices, and have high overhead, implementations are not well optimized, do not take advantage of auxiliary information, and do not gracefully tolerate soft-faults which modern architectures introduce. Our dynamic approaches for PageRank and community detection address these issues. Our work has been accepted by IPDPS workshops (3), the Euro-Par conference (1), the ICPP conference (1), and the Complex Networks conference (1). Key outputs from our work include the design of a common framework for dynamic graph algorithms, and techniques to address soft faults in dynamic algorithms.


Publications


Technical Reports


Manuscripts


Thesis Materials


Software

Tool Description
📦 nvgraph.sh CLI for nvGraph, which is a GPU-based graph analytics library written by NVIDIA, using CUDA.
📦 snap-data.sh CLI for SNAP dataset, which is a collection of more than 50 large networks.
⛏️ graph-properties List a few graph properties.
⛏️ graph-generate Perform certain operations upon a fixed graph.
🧵 graphs A few sample graphs in Matrix Market (.mtx) format.


Others