Big Data Analytics Using FPGAs
Large modern FPGAs provide more than 3000 dedicated multiplier blocks that can run at a speed of up to 500 MHz and the aggregated internal memory throughput can reach many terabytes per second. In the APT group, we are currently building an FPGA-based big data analytics machine that will harness the compute power of 2 large FPGAs, the aggregated throughput and capacity of 48 fast SSDs and a networking speed of 8 10Gb optical interconnects in a desktop form factor.
The goal of the project is to develop big data analytics applications for our new data analytics machine. The exact algorithms/application can be self-defined or will be given. Tasks may include, for example, histogram computation, text analytics, data clustering and/or data ranking and/or skyline queries (i.e. computing a Pareto-optimal data set).