Real-time Spread Burst Detection in Data Streaming
Real-time Spread Burst Detection in Data Streaming
复制标题
DOI：
10.1145/3589979
复制
发表时间：
2023-05
期刊：
影响因子：
--
通讯作者：
Haibo Wang;Dimitrios Melissourgos;Chaoyi Ma;Shigang Chen
中科院分区：
文献类型：
--
作者：
Haibo Wang;Dimitrios Melissourgos;Chaoyi Ma;Shigang Chen
数据流在网络监控、Web服务、电子商务、股票交易、社交网络和分布式传感等领域有着广泛的应用。本文提出了一种在流量大小上不同于传统突发检测的新问题--流量扩散中的实时突发检测问题。它在网络安全、网络工程、互联网趋势识别等方面具有重要的实际应用价值。这是一个具有挑战性的问题，因为估计流量扩散需要我们记住所有过去的数据项，而实时检测突发需要我们将扩散估计开销最小化，而这在大多数以前的工作中并不是优先考虑的。为扩频突发检测提供了第一个高效、实时的解决方案。它是基于一种新的实时超级撒布器识别器设计的，在准确性和处理开销方面都超过了最先进的水平。超级散布器识别器反过来又基于一种新的草图设计，用于实时扩散估计，其性能优于现有的最佳草图。
Data streaming has many applications in network monitoring, web services, e-commerce, stock trading, social networks, and distributed sensing. This paper introduces a new problem of real-time burst detection in flow spread, which differs from the traditional problem of burst detection in flow size. It is practically significant with potential applications in cybersecurity, network engineering, and trend identification on the Internet. It is a challenging problem because estimating flow spread requires us to remember all past data items and detecting bursts in real time requires us to minimize spread estimation overhead, which was not the priority in most prior work. This paper provides the first efficient, real-time solution for spread burst detection. It is designed based on a new real-time super spreader identifier, which outperforms the state of the art in terms of both accuracy and processing overhead. The super spreader identifier is in turn based on a new sketch design for real-time spread estimation, which outperforms the best existing sketches.