Publication at Faculty of Mathematics and Physics |
2015
Abstract
Data streaming systems have been successfully employed for various data processing tasks. Their main benefit is that they simplify the design of data-intensive applications and they introduce many opportunities for task, data, and pipeline parallelism.
In this work, we are proposing an enhancement for data streaming systems that allows distributed processing of the data streams and also introduce parallel accelerators, which can be utilized for data parallel subtasks. The viability of our approach is verified by integrating the support for heterogeneous accelerators into the Bobox system, which is a parallel framework for data stream processing.