Computer architects and database system designers have long enjoyed fruitful interactions both at research and production environments. Database workloads are crucial in evaluating server chip performance, and the arrival of new CPU features routinely initiates new investigations on architecture-conscious software design. Yet, despite several research proposals on hardware-software co-design and (30-year old) talks on "database machines," general-purpose hardware still dominates in today's data centers.
There is increasing evidence, however, that hardware specialization and diversification may be necessary in the not too distant future. On the software side, unprecedented volumes of data and new, internet-scale user-facing applications have challenged system designers and led to the dismantling of the "one-size-fits-all" software design mantra. On the hardware side, energy consumption concerns significantly affect new chip design; at the same time, a re-haul of memory hierarchy technologies is well underway, led by the introduction of new non-volatile RAM technologies. Some high-level forms of specialization and hardware-software co-design have already appeared as high-performance database "appliance" boxes, "wimpy-node" energy-efficient architectures, GPU-assisted platforms, and FPGA-based systems.
In this panel, we ask our panelists, a mix of computer architects and database system designers, what are the current state, potential, opportunities and caveats in designing future data-centric systems under a joint hardware-software approach. Some of the topics that we discuss in the panel are: recent developments in hardware and software design, upcoming and future memory technologies, hybrid chip/data center designs, best practices for energy efficiency, the role of FPGAs, GPUs, and non-server hardware in data management platforms, along with the research community's potential for impact in next-generation data-center design and deployment.