III: Medium: Longview: Querying the Future Now

点击翻译按钮获取中文摘要

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).Traditional database systems are designed to provide the user with aclear picture of past states of the world as is represented in thedatabase. Recently, stream processing systems are introduced toproduce near real-time answers for those applications applicationsthat require up-to-date information. Thus, we see a trend towardshrinking the ?reality gap? to zero. But for some applications, evenreal-time is not good enough. There is often a desire to get out infront of the present by delivering predictions of future events totake advantage of opportunities or to avert calamity. Securityapplications are a good example since they are typically interested inpreventing a breach rather than simply reporting that one hashappened. There is currently no database system that can effectivelyserve as a generic platform to support such predictive applications.This project aims to fill this gap by designing and building aprototype database system called "Longview" to enable data-centricpredictive analytics. Longview facilitates the use of statisticalmodels to analyze historical and current data and make predictionsabout future data values and events. Users can plug new predictivemodels into the system along with a modest amount of meta-data, andthe system uses those models to efficiently evaluate predictivequeries.Longview treats predictive models as first-class citizens byintelligently managing them in the process of data management andquery optimization. This involves automatically building models anddetermining when and which model(s) to apply to answer predictivequeries. This also involves creating and using the proper physicaldata structures to facilitate efficient model building, selection, andexecution. Longview handles both streaming and historical queries. Infact, many streaming queries need efficient access to an archive ofpast values, making it necessary to seamlessly combine both stream andhistorical processing. Finally, Longview investigates "white-box"model support, in which the database leverages the operationalsemantics and representation of models to improve performance.Longview's goal is to make it much easier to build predictiveanalytics applications in data intensive situations. Seamlesslycombining data and model management is key to make the process ofcomputing with predictions far easier to express and more efficientthan the current ad-hoc application-level approaches. The resultingtechnology also allows for a better understanding and support foruser-defined functions in database systems.Longview is initially used for a real-world sensor-based trackingapplication and a predictive web portal for easy experimentation withdifferent models and data sets. Further information on the project canbe found on the project web page: http://database.cs.brown.edu/projects/longview/