Improving Responsiveness for Wide-Area Data Access

In a wide-area environment, the time required to obtain data from remote sources can vary unpredictably due to network congestion, link failure or other problems. Traditional techniques for query optimization and query execution do not cope well with such unpredictability. The static nature of those techniques prevents them from adapting to remote access delays that arise at runtime. In this paper we describe two separate, but related techniques aimed at tackling this problem. The first technique, called Query Scrambling, hides relatively short, intermittent delays by dynamically adjusting query execution plans on-the-fly. The second technique addresses the longer-term unavailability of data sources by allowing the return of partial query answers when some of the data needed to fully answer a query are missing.