Abstract
World Wide Web content providers often resort to "cache-busting" in order to obtain demographic information. Object usage reporting methods have been proposed to address this problem. We quantitatively compare strategies for reporting object hits from proxy caches back to origin servers, and propose novel strategies for improving reporting performance and efficiency. Examining hit-metering and usage-limiting approaches proposed in RFC 2227, we find a fundamental trade-off between reporting latency and efficiency. Further, we find the temporal locality in the server reference stream to be significantly stronger than that in the object reference stream. We propose a server report aggregation strategy that leverages this fact, and show that it can reduce reporting latency and improve efficiency by as much as 80% and 100% respectively. We also propose and evaluate additional strategies to improve performance. These include: dynamic reporting thresholds, report aggregation in a cache hierarchy, and piggybacking reports on existing HTTP messages.