Research And Implementation Of Network Quality Bottleneck Analysis Method In IDC Coverage Area

The analysis of network quality bottlenecks in data center coverage area networks is an emerging field in the domain of cloud network congestion detection and localization,attracting significant attention in the industry.Microsoft has previously employed a combination of passive and active measurements to diagnose network issues for users in the coverage areas of their data centers.However,this approach has stringent prerequisites that make it challenging for other cloud service providers to replicate.Jitterbug employs an active measurement approach to collect round-trip time(RTT)and jitter metrics for router interface IPs in close proximity to the measurement source.It utilizes anomaly detection to infer congestion but is limited in its applicability to inter-AS links near the measurement points,lacking generality.This paper addresses the challenges of inferring network congestion in data center coverage area networks through active measurements.However,it encounters several complex issues.First,network paths in data center coverage areas cannot guarantee a consistent response to measurements,and the presence of unknown nodes introduces noise into the paths.Second,network paths in data center coverage areas have multiple hops,traverse various ASes,ISPs,and countries,resulting in inconsistent measurement frequencies and durations for the combinations of paths and router interface IPs.This variability creates a lack of effective congestion inference solutions for path segments formed by path and router interface IP combinations.To address these issues,this paper introduces an approach for analyzing network quality bottlenecks in data center coverage area networks.This approach combines time-series congestion inference methods with statistical-based congestion inference methods to analyze congestion in path segments formed by path and router interface IP combinations.It discusses the characteristics and distribution of network quality bottlenecks in data center coverage area networks.The paper’s main research contributions are summarized as follows:(1)Implemented an IP path analysis solution tailored for data center coverage area networks.Based on the characteristics of network measurements in data center coverage area networks,this paper introduces metrics such as path persistence and regional coverage rate to assess the feasibility of analyzing the network quality bottleneck in data center coverage area networks.The chosen granularity for data center network quality bottleneck analysis is AS-GEO Path,where each node in the path is represented by the corresponding AS number and geographic coordinates of the IP address.(2)Based on the universal optimization of parameter settings in Jitterbug,proposed a framework called Universal Jitterbug for inferring path congestion from a time series perspective.Building upon the existing time series congestion inference framework Jitterbug,improvements are made to the parameter acquisition and calculation methods.The calculation of the minimum RTT time series window has been changed from a fixed 15 minutes to half of the congestion duration to ensure that the probability of measuring RTT within the time window is greater than or equal to 99%.The time window for calculating the quartile range time series of jitter has been adjusted from a fixed 150 minutes to 30 times the time window for calculating the minimum RTT time series.The threshold for anomaly detection has been set to the 95th percentile of historical statistical data.The congestion inference of data center coverage area network paths and router interface IP combinations is achieved using both latency and jitter indicators.(3)Proposed a statistical-based approach for analyzing network quality bottlenecks in data center coverage area networks.For scenarios where congestion inference based on time series data is not feasible,statistical metrics,including percentiles,mean,and median,are used to describe the characteristics of the RTT measurement data for paths.Thresholds are obtained from historical measurement data to assess the congestion status of current paths,providing a statistical approach for congestion inference in data center coverage area networks.(4)Developed a data center coverage area network bottleneck analysis system suitable for the scope of this research.The paper defines system development requirements,designs the system architecture,describes the functions and implementations of various system modules,and conducts testing.