BGP路由传播异常特征数据库

本文介绍BGP路由传播异常特征数据库，面向路由异常检测、传播分析与泄露评估提供可复现的数据基础。数据库由四个互补子数据集组成：(1) 基于SCREEP平台采集的仿真网络多模态日志数据，覆盖100个节点（50个AS、50个Host），采集于2025年10–11月，数据量245GB，JSON格式，包含SystemNetwork、BGP、Router、Ping、Topology、Attack六类日志，用于刻画资源状态、协议交互与拓扑变化的跨层演化；(2) 仿真网络时序指标数据，采集于2025年10–11月，包含8个AS节点，CSV格式，总量87MB，划分为8天正常训练集与12天随机攻击验证集并给出异常标注；(3) 真实BGP异常事件数据集，覆盖2004–2024年共38个经人工验证的事件（18起劫持、12起泄漏、8起中断），约300GB原始BGP UPDATE消息、超过40亿条更新记录，支持点级与事件级评测；(4) 面向概率推理与路径评分的路由泄露数据资源集合，包含RIB_Year（2024年8月至2025年7月每月1–5日）与RIB_Day（2025年7月1–10日、8小时粒度）AS路径数据、AS关系概率分布数据，以及基于Cloudflare Radar（2025年6月1–30日）构建的泄露/合法路径级标注数据。四类数据均采用统一清洗与质量控制流程，并提供可追溯元数据与标准化组织，便于算法开发、对比评测与工程验证。This paper introduces the BGP Routing Anomaly Propagation Feature Database, which provides a reproducible data foundation for routing anomaly detection, propagation analysis, and leakage evaluation. The database consists of four complementary sub-datasets: (1) Multimodal log data of simulated networks collected based on the SCREEP platform, covering 100 nodes (50 Autonomous Systems (ASes) and 50 hosts). Collected from October to November 2025, the dataset totals 245GB in JSON format, and includes six types of logs: SystemNetwork, BGP, Router, Ping, Topology, and Attack. It is used to characterize the cross-layer evolution of resource states, protocol interactions, and topology changes. (2) Time-series indicator data of simulated networks, collected from October to November 2025, covering 8 AS nodes. Stored in CSV format with a total size of 87MB, the dataset is split into an 8-day normal training set and a 12-day random attack validation set, with anomaly labels provided. (3) Real-world BGP anomaly event dataset, covering 38 manually verified events from 2004 to 2024 (18 hijacks, 12 leaks, 8 outages). It contains approximately 300GB of raw BGP UPDATE messages and over 4 billion update records, supporting both node-level and event-level evaluation. (4) Routing leakage data resource collection for probabilistic inference and path scoring, including AS path data of RIB_Year (1st to 5th of each month from August 2024 to July 2025) and RIB_Day (1st to 10th July 2025, 8-hour granularity), AS relationship probability distribution data, and leakage/legitimate path-level annotation data constructed based on Cloudflare Radar (June 1st to 30th, 2025). All four datasets adopt a unified cleaning and quality control workflow, and provide traceable metadata and standardized organization, facilitating algorithm development, comparative evaluation, and engineering validation.

搜集汇总

背景与挑战

背景概述

BGP路由传播异常特征数据库是一个面向路由异常检测、传播分析与泄露评估的数据资源，由仿真网络多模态日志数据、时序指标数据、真实BGP异常事件数据集和路由泄露数据资源集合四个互补子集构成。所有数据均经过统一清洗与质量控制，提供可追溯元数据和标准化组织，支持算法开发、对比评测与工程验证。