This paper presents a comprehensive study of Internet Web server workloads, focusing on six different data sets from various environments: three from academic institutions, two from scientific research organizations, and one from a commercial Internet provider. The study aims to identify workload invariants that are universal across all Internet Web servers. Ten invariants are identified, including the high frequency of successful responses, the dominance of HTML and image documents, the small average size of transferred documents, the concentration of references to a small number of documents, and the temporal and geographic distribution of requests. The paper also discusses the implications of these invariants for caching and performance improvements, highlighting the trade-offs between reducing the number of requests and the volume of network traffic. The findings suggest that caching strategies can significantly enhance Web server performance, particularly for small files, while also addressing the challenges posed by large files and one-time document references.This paper presents a comprehensive study of Internet Web server workloads, focusing on six different data sets from various environments: three from academic institutions, two from scientific research organizations, and one from a commercial Internet provider. The study aims to identify workload invariants that are universal across all Internet Web servers. Ten invariants are identified, including the high frequency of successful responses, the dominance of HTML and image documents, the small average size of transferred documents, the concentration of references to a small number of documents, and the temporal and geographic distribution of requests. The paper also discusses the implications of these invariants for caching and performance improvements, highlighting the trade-offs between reducing the number of requests and the volume of network traffic. The findings suggest that caching strategies can significantly enhance Web server performance, particularly for small files, while also addressing the challenges posed by large files and one-time document references.