Automated composition and execution of hardware-accelerated operator graphs
Automated composition and execution of hardware-accelerated operator graphs
复制标题
硬件加速操作图的自动组成和执行
DOI:
10.1109/recosoc.2015.7238078
复制
发表时间:
2015
期刊:
影响因子:
--
通讯作者:
Thilo Pionteck
中科院分区:
文献类型:
--
作者:
Stefan Werner;Dennis Heinrich;J. Piper;Sven Groppe;Rico Backasch;Christopher Blochwitz;Thilo Pionteck
In this paper, we present the fully automated composition and execution of Semantic Web queries within a hardware/software system which uses a Field Programmable Gate Array (FPGA) as an accelerator. The presented approach allows to write a query in the database's front-end, transparently executes all steps to retrieve a configuration suitable for the FPGA which represents the given query, and obtains the result by evaluating the query on the FPGA. In order to obtain a runtime-reconfigurable framework we define a static and a dynamic partition. The dynamic partition itself is composed by using a predefined query template and a pool of operators. The operators of the given query are written automatically into the template and connected accordingly. After reconfiguration of the FPGA the host system supplies the initial data to the FPGA which computes the final result and sends it back to the host system to be displayed to the user or application. The evaluation shows that not all queries may take benefit from a dedicated hardware-accelerator but shows promising speedup for complex queries.