Synopsis based load shedding in XML streams

Stream systems are susceptible to variations in data arrival rate. At times, data arrival rate may spike up to cause unacceptable output latencies and unpredictable system behavior. Recently, load shedding systems have been proposed to deal with this situation. But almost all these systems are for relational data streams and, to the best of our knowledge, none has been proposed for XML data streams so far except [15]. Dropping data randomly may have been an effective method for load shedding in the relational context, due to the uniformity of relational data. But in the XML context, the same method will lead to much invasive negative effect on processing of XML queries due to the recursive and nested structure of XML data. We propose a load shedding framework for XML data streams. We explore the effectiveness of various load shedding techniques based on a general load shedding strategy that takes into account QoS parameters and relative accuracy of the query results. We implement various load shedding strategies and present their result.