Abstrakt
Current hardware development trends exhibit clear inclination towards parallelism. Multicore CPUs as well as many-core architectures such as GPUs or Xeon Phi devices are widely present in both high-end servers and common desktop PCs.
In order to utilize the computational power of these parallel platforms, the applications must be designed in a way that intensively exploits concurrent processing. In our work, we propose techniques that simplify the application decomposition process in data streaming systems.
The data streaming paradigm may be applied in many data-intensive applications, e.g., database management systems or scientific data processing. In order to employ these techniques, we have developed a data streaming language called Bobolang that simplifies the design of the application.
This approach allows the programmer to write strictly serial operators in a traditional language and then interconnect these operators in an execution plan, that presents opportunities for automated parallel processing.