A New Linear Programming Based Load-Shedding Strategy

Abstract

There is a large class of applications that high-frequency data produced continuously in 7 days per week and 24 hours per day. The traditional data processing system can not deal with them efficiently because of the active data pushing and the passive queries. With the purpose of dealing with these data, data stream management systems (DSMS) appeared. For the geographical distribution of data streams, distributed data stream processing systems are studied recently. As an important aspect of distributed data stream management system (DDSMS), load management can balance system load incurred by unpredictable incoming data stream and inappropriate query operators' distribution. Load-shedding acts as an important role in load management of distributed data stream system. In this paper, a new linear programming based load-shedding strategy is proposed to degrade system load on a simple query network with network constraint. Compared with previous work which concentrates on one factor only, it takes two factors, include CPU capacity and network constraints. Experiments indicate that it can increase system stability and to decrease the loss of query accuracy with low extra cost.