Dynamic operator splitting in mobile CEP scenarios

Abstract

With an increasing number of mobile devices being used in modern day to day life, applications using data provided by their environment are becoming more and more common. Complex Event Processing systems are a popular approach to realize them. Usually the data, collected for example by the mobile devices' built in sensors like GPS is sent to a stationary host for evaluation. Since distances between host and data source might be quite large, this is usually done by using communication standards like 3G or GSM. Since energy costs of such messages are rather high it is in some cases more energy efficient to evaluate some data directly on the mobile device before sending it, in order to reduce communication. In large scale networks however, this will still result in a lot of expensive 3G or GSM messages being sent, since each mobile device will have to send its data eventually. This work will introduce algorithms to create a simple network structure that takes advantage of the mobile device's locality in order to create clusters of mobile devices. In these clusters, members use the much cheaper wireless lan communication to collect data, which will then be pre-processed locally before being sent to the host via the more expensive 3G or GSM for final processing. To make pre-processing possible we will also examine Operators used in CEP systems in terms of their ability to be distributed further, and provide a classification based on this. Finally we will evaluate our system by performing simulations. We will test the general performance of our algorithms by measuring the energy consumed to transmit events. Our evaluation shows that our approach is able to safe energy for systems that produce events with a high frequency.