Tech Report CS-04-13

Window-aware Load Shedding for Data Streams

Nesime Tatbul and Stan Zdonik

September 2004


Data stream management systems may be subject to higher input rates than their resources can handle. In this case, results get delayed and Quality of Service (QoS) at system outputs may fall below acceptable levels. Load shedding addresses this problem by allowing data loss in exchange for reduced latency. Drop operators are placed at carefully chosen points in a query plan, in order to relieve overload with minimal loss in answer quality. In this paper, we describe a load shedding technique for queries consisting of one or more aggregate operators with sliding windows. We introduce a sophisticated drop operator, called a "Windowed Drop". This operator is aware of window properties (i.e., window size and window slide) of downstream aggregate operators in the query plan. Accordingly, it logically partitions the stream into windows and probabilistically decides which windows to drop. This decision is further encoded into tuples by marking the ones that are disallowed from starting new windows. Unlike earlier approaches, our approach preserves integrity of windows throughout a query plan, and always delivers subsets of original query answers with minimal degradation in overall QoS utility.

(complete text in pdf or gzipped postscript)