Real-time analytics on Big Data
The unlimited increase in data volume and the need to interpret this data in real-time to detect complex situations have motivated research communities to develop technologies that process high-volume data on-the-fly. Complex Event Processing (CEP) is an effective method to process such event streams to detect complex patterns being of interest to applications. This project aims at development of a CEP framework that incorporates methods for the cost-efficient configuration and dynamic adaptation of CEP systems to meet latency requirements of applications. Moreover, in the presence of resource limitation, the focus is on approximation or load shedding methods for operators and their impact on the latency and quality of CEP processing. The overall goal is to develop load shedding methods that ensure a given latency bound for an operator graph, while maximizing the quality of result perceived by the application.
Distributed Graph Processing
Graphs are ubiquitous in the modern world - the internet, power supply system, route planning, and many more applications are built on top of them. The growing volume of graph-structured data is immense and, in today’s world, these data sets are already into trillion scale. Processing such huge amount of data takes much time and computational resources. Moreover, we are faced with the trend that the growth rate of data outstrips the growth rate of processors. As a result, in this project, we work towards fast, scalable, and resource efficient graph processing applications. To do so, we explore new optimized partitioning strategies to distribute graphs across multiple machines efficiently. Furthermore, we investigate the possibilities of approximate computing in graph processing to run graph algorithms orders of magnitude faster, focusing on finding good trade-offs between runtime improvement and loss of quality.
The project aims at exploiting the recent advances in software-defined networking (SDN) technologies to enable highly efficient communication middleware systems (e.g., publish/subscribe systems) and networked applications (e.g., networked control systems as used in automation). To this end, the project addresses challenges associated with both the data plane and the control plane. Concepts are developed to push the functionality - that was previously implemented on the application layer of the OSI model - to the hardware switches (in the data plane) to substantially improve the performance of middleware/application in terms of latency and throughput.