neoDBMS2: Hardware/Software Co-Design for Update-Capable NDP-Accelerated Databases on cache-coherently attached Scalable Computational Storage

• Applicants: Professor Dr.-Ing. Andreas Koch; Professor Dr.-Ing. Ilia Petrov
• Subject Area: Computer Architecture, Embedded and Massively Parallel Systems ; Security and Dependability, Operating-, Communication- and Distributed Systems
• Term: since 2019
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 419942270

neoDBMS2 is an extension proposal that builds on the successful results of the first project phase. With the advent of cache-coherent interconnects such as CXL, neoDBMS2 pursues architectures for larger-than-memory, NDP-capable, post-Moore DBMS on heterogeneous systems. To this end, neoDBMS aims at investigating the following core research challenges. Firstly, investigate approaches allowing the computational storage to perform modifications (updates) as hardware-accelerated NDP, while remaining transactionally consistent with any host-side DBMS updates. In this context, we will investigate new lightweight synchronization mechanisms that can be realized over a modern cache-coherent interconnect protocol such as CXL.cache and logging and recovery approaches. Secondly, neoDBMS2 will address a new type of systems that are a hybrid between more performant scale-up and distributed scale-out systems. We aim at better scale-out our architecture for larger data volumes and throughputs, and address multi-NDP-device systems. Thirdly, neoDBMS2 pursues investigating heterogeneous parallel execution models that can execute flexibly across the hardware/software boundary for NDP-accelerated system. Furthermore, we aim at finely granular cooperative execution models that allow NDP results to be handed back to the host DBMS as soon as they are available and thus interleaving DBMS and NDP processing. Lastly, abstractions and adapted workhorse data structures specialized for this hardware/software architecture will be investigated, as they will need to be able to efficiently support simultaneous access from the host engine and CPUs, and the NDP operations, running on heterogeneous on-device hardware with different memory-hierarchies and cost-models.

• DFG Programme: Research Grants