Flashy-DB: Impact of Flash Solid State Disks on Performance and Architecture of Data-Intensive Software Systems

• Applicants: Professor Dr. Alejandro P. Buchmann; Professor Dr.-Ing. Ilia Petrov
• Subject Area: Software Engineering and Programming Languages
• Term: from 2010 to 2015
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 165012865

Flash Memories with radically different properties are an evolving storage technology. Existing concurrency control approaches, optimization techniques, buffer management strategies, access paths and in essence the whole DBMS architecture are designed to compensate for traditional storage media (hard disks). To make full use of the characteristics of evolving storage technologies (such as Flash-SSDs), the existing architectural fundamentals need to be critically reevaluated. Since Flash-SSDs will have different impact on Online Transaction Processing (OLTP) and Online Analytical Processing (OLAP) systems, we propose to study the impact of flash-SSDs on OLTP systems as well as data warehouses.This proposal is a renewal proposal. In the first phase we concentrated primarily on exploiting Flash-SSDs in OLTP systems. The research conducted in the first two years of the project resulted already in 9 accepted and 3 submitted publications, 3 completed Master theses (2 in progress) and 1 completed Bachelor Thesis (2 in progress). The main research directions we plan to explore are multi-versioning in database systems, the direct use of FTL-less flash storage and changing memory hierarchies. Firstly, a novel aspect of Multi- Versioning (MV) approaches is that they can be used to address weaknesses of new storage technologies such as asymmetry and endurance, since MV techniques can result in out-of-place update techniques. Secondly, multi-versioning can be effectively used to address changing memory hierarchy and data placement. Thirdly, the use of Flash Translation Layer techniques as block-device compatibility layer promotes easy deployment, but incurs a major performance penalty and significant complexity. Database systems have rich page information as well as I/O pattern knowledge, which can enable direct, FTL-less use of Flash storage to achieve better performance and predictability.

• DFG Programme: Research Grants