This Athene Young Investigator award supports a research plan to develop performance evaluation methods for parallelized communication systems under synchronization constraints.
Modern communication networks rely heavily on parallel multi-server systems, e.g., for multipath transmission protocols, web server farms, networked high performance computing systems, as well as, real-time data analytics. These systems exploit parallelization to provide lower latency, capacity scalability and higher reliability. Although accurate performance models for parallel communication systems are essential to guide the architectural design of the future Internet, understanding the performance properties of such systems remains, however, notoriously hard. A particular difficulty arises since such systems naturally comprise synchronization events due to the intrinsic modes of operation of many protocols and applications - take for example the result aggregation in big data analysis systems such as MapReduce /Hadoop or the in-order output of Multipath TCP.
The work within this project builds on the frameworks of queuing theory and stochastic network calculus to provide stochastic bounds on the performance of parallelized systems in terms of throughput and delay distributions. The methods developed here will enable analytical investigations of network protocols and applications that actively control multi-server architectures under synchronization. The developed models will directly contribute to the optimization of applications such as adaptive, multipath-aware video streaming algorithms, as well as, into the design of scheduling and routing algorithms for future Internet architectures.