Improving Data Delivery in Wide Area and Mobile Environments

Abstract

The popularity of the Internet has dramatically increased the diversity of clients and applications that access data across wide area networks and mobile environments. Data delivery in these environments presents several challenges. First, applications often have diverse requirements with respect to the latency of their requests and recency of data. Traditional data delivery architectures do not provide interfaces to express these requirements. Second, it is difficult to accurately estimate when objects are updated. Existing solutions either require servers to notify clients (push-based), which adds overhead at servers and may not scale, or require clients to contact servers (pull-based), which rely on estimates that are often inaccurate in practice. Third, cache managers need a flexible and scalable way to determine if an object in the cache meets a client's latency and recency preferences. Finally, mobile clients who access data on wireless networks share limited wireless bandwidth and typically have different QoS requirements for different applications. In this dissertation we address these challenges using two complementary techniques, client profiles and server cooperation. Client profiles are a set of parameters that enable clients to communicate application-specific latency and recency preferences to caches and wireless base stations. Profiles are used by cache managers to determine whether to deliver a cached object to the client or to validate the object at a remote server, and for scheduling data delivery to mobile clients. Server cooperation enables servers to provide resource information to cache managers, which enables cache managers to estimate the recency of cached objects. The main contributions of this dissertation are as follows: First, we present a flexible and scalable architecture to support client profiles that is straightforward to implement at a cache. wireless base station. Second, we present techniques to improve estimates of the recency of cached objects using server cooperation by increasing the amount of information servers provide to caches. Third, for mobile clients, we present a framework for incorporating profiles into the cache utilization, downloading, and scheduling decisions at a We evaluate client profiles and server cooperation using synthetic and trace data. Finally, we present an implementation of profiles and experimental results.