Serverless Data Management Primitives for Software-defined Composable Systems

点击翻译按钮获取中文摘要

The scientific as well as the commercial domain of data management has experienced a significant push within the last two decades. "Data as the new oil" has sparked research and innovation in all areas of data management systems, from theory via system-related work to data-intensive applications. In addition to novel data management architectures, the research community also explored the efficient use of modern hardware like GPU, FPGA, RDMA, etc.Within this project however, we plan to take the next step and target an upcoming hardware technology currently discussed in the context of Composable Systems. Composable Systems are software-defined computers directly "soft-wiring" a collection of real hardware components and thus completely blurring the boundaries of traditional Scale-Out or Scale-Up architectures. Since such a "Logical" system may be dynamically re-composed, each algorithm may define the "optimal" composition by instantly adding/removing computing units, changing memory and storage assignments, and modifying the topology of the memory and data backplane. While software is traditionally striving to optimally exploit the use of existing (hard-wired) hardware, the goal is now to identify the optimal combination of a software and (soft-wired) hardware configuration. The characteristics of such software-defined infrastructures require a fundamental re-thinking of many existing algorithms and data structures. Being able to re-compose many aspects of an underlying "Computer" during runtime (!) of an application opens up a large field of widely unexplored research opportunities. Reflecting the nature of a Reinhart Koselleck proposal, we aim at investigating the impact of such "next-next"-generation hardware environments of Composable Systems on the design and architecture of data management primitives to efficiently run large-scale but also transactionally correct query workloads. As of now, we are not aware of any dedicated data management research project with this particular focus on Composable Systems. We strive to produce internationally visible results and expect publications and presentations at international conferences like ACM SIGMOD, VLDB (PVLDB), IEEE ICDE as well as within journals like ACM TODS, VLDB Journal, and IEEE Data Engineering Bulletin.