zur Startseite

PRECEPT Research Project

Parallel complex event processing to meet probabilistic latency bounds
ProjekttypResearch Project
Gefördert durch DFG (SFB 467)
Beginn 2015/01/01
Leiter Prof. Dr. rer. nat. Dr. h. c. Kurt Rothermel
Prof. Dr. Umakishore Ramachandran
Mitarbeiter Tariq, Muhammad Adnan
Mayer, Ruben
Slo, Ahmad
Ansprechpartner Tariq, Muhammad Adnan
Mayer, Ruben
Kooperationspartner Georgia Institute of Technology (Georgia Tech)

Nowadays, billions of sensors and smart objects, i.e., objects with embedded electronics that enable identification, sensing and actuation capabilities, are deployed throughout the globe. These sensors and smart objects---e.g., smart meters, GPS sensors and RFID tags---continuously collect data about the physical world. The captured data streams promise to pave the way towards the Internet of Things (IoT) by enabling modern applications in the areas of smart homes, smart cities, environmental monitoring, healthcare etc., to detect and react to live-situations in the surrounding world.

To this end, Complex Event Processing (CEP) systems offer means to efficiently detect event patterns, or complex events, from the sensor streams that correspond to situations of interest to the smart applications. This way, CEP systems help in realizing a distributed intelligence in the IoT. However, in many cases consistent and timely detection of situations is crucial.  Consistency in this regard means that neither false positives nor false negatives should be detected. The situations that are detected inconsistently or late reduce the benefit an application can gain from the CEP system. For instance, taking a turn without receiving an update about traffic congestion or reacting late to the occurrence of such a situation restricts a navigation application in its ability to find optimal traffic routes.

With the increasing number of data sources (i.e., sensors) and the increasing volume at which data is produced, parallelization of situation detection is necessary to speed up CEP systems such that they are capable of meeting application-defined latency bounds at minimal cost and can ensure the consistency of produced event streams. In summary, this project aims to make the following research contributions:

1) Develop methods for fine-grained data parallelization by establishing a pattern sensitive fission model. This model will allow the dynamic adaptation of the parallelization degree, supporting high and dynamic event rates.

2) Develop methods for the configuration and dynamic adaptation of CEP processing units called operators so that parallel operators are assigned to resources of a data center and they can probabilistically meet a latency bound under varying and dynamic workloads.

3) Develop methods for the cost-minimal configurations of a network of CEP operators so that the entire network of operators can meet probabilistic end-to-end latency bounds.

4) Evaluate the proposed concepts and methods using real-world workloads of camera networks and traffic monitoring.