Reconfigurable Database Acceleration

The data volume in large databases can easily exceed many terabytes of data and processing this data is becoming very difficult. For example, efficient (sequential) sorting algorithms have a time complexity of n log n and even the log n factor is an issue when sorting terabyte problems. However, there exist parallel sorting accelerators on FPGAs (Field Programmable Gate Arrays) that take log n sorter units to sort in linear time.

FPGAs are a class of powerful massively parallel processing devices that shall be used in this PhD project for substantially accelerating data warehouse databases or data-stream management systems. The idea is to build a library of, for example, SQL operator modules that will be dynamically stitched together to accelerate individual database queries. By using wide data paths and a high level of parallel processing, processing speed of one to three orders of magnitude is expected while consuming significantly less power than a server CPU.