A distributed streaming framework for edge–cloud triangle counting in graph streams

The triangle counting problem in graph streams has been extensively studied in social network analysis, recommendation systems, user portraits and other fields. However, cloud computing based streaming algorithms cause high bandwidth occupation and long transmission latency due to limited bandwidth of the cloud. Recently, edge computing is promising to overcome the issue of transmitting large-scale data for cloud computing. However, directly applying edge computing in streaming triangle counting will reduce the accuracy of the triangle count estimation, due to the limitation of local computing at the edge network. We term the cooperations between edge computing and cloud computing for streaming triangle counting as edge–cloud triangle counting in graph streams. In this paper, we first propose a streaming framework for edge–cloud triangle counting in graph streams. Then, we propose a streaming triangle counting algorithm called Trie-based Edge Compression (TbEC) by using the binary trie at the edge network that enables lossless compression and efficient transmission to the cloud. In addition, to extend our algorithms for triangle counting in multigraphs, we present a dual deduplication strategy collaboratively using the trie-based data structure and a Bloom Filter. Our experiments with real-world datasets show that TbEC is (a) Accurate: yielding up to 3.35 × more accurate smaller estimation error than the state-of-the-art distributed streaming algorithm, (b) Fast: yielding up to 10.59 × faster than the state-of-the-art distributed streaming algorithm, (c) Scalable: scaling linearly with the number of edges in the input graph stream.