Abstract

As databases need to strike a good balance of high performance at low costs, hardware-accelerators like FPGAs are increasingly considered to improve core database operations. In this paper we complement previous research, where hardware accelerators are tightly integrated with the host CPU, and expose the FPGA as a hardware data processing service to a cluster of databases to delegate the asynchronous execution of compute-intensive database operations. We apply our novel architecture, called DASH, to the use case of string dictionary compression using the compute-intensive RePair compression and demonstrate its benefit in this distributed architecture. We also explain how DASH can be applied to other database scenarios and propose a research agenda.