Abstract

As the norm in networking shifts towards adopting encrypted traffic, content delivery networks (CDNs) struggle with profiling their performance metrics due to a lack of visibility into packets' contents. This research provides a complete solution for the problem of encrypted traffic pattern analysis in CDNs with focus on performance profiling without payload inspection. The framework proposed here employs traffic flow characteristics, traffic, time series analysis, and TLS handshake parameters to determine critical performance indicators which include but not limited to: latency and throughput bottlenecks, retransmission rates, delivery efficiency, and others. The performance of the Content Delivery Networks (CDN) and their performance under varying loads and different encryption protocols were estimated through classification and regression modeling techniques which incorporated machine learning tools. Data from different locations were integrated and processed using a multi-layered profiling pipeline which consists of feature extraction, anomaly detection, and predictive modeling. The evaluation shows that performance profiling is achievable using flow-level metadata alone, even with encryption restrictions. In addition, the results indicate that CDN performance can be reliably estimated from the temporal patterns of the packets and the inter-chronal intervals of their arrivals. This study contributes to the emerging domain of encrypted traffic analytics and can be used by service providers, researchers, or developers to enhance CDN strategies for content delivery without violating the privacy or confidentiality of user data.