Within the DFG-funded project “Architecture of future transaction systems”(1990-1993), PI Erhard Rahm, we have evaluated the use of extended storage hierarchies. The focus was on the use of page-addressable and possibly non-volatile semiconductor stores located between main memory and disk. Three storage types are considered: extended memory, solid-state disks and disk caches. These storage types are used either as a second-level database buffer, as a write buffer or to permanently store entire files (e.g., the log file).

In the distributed (closely coupled) case further opportunities exist, in particular in Shared Disk systems to support global concurrency control and coherency control, global logging, transaction routing and to speed-up inter-system communication.

New caching strategies for both main memory and extended memory have been developed for speeding-up the processing of batch jobs in order to maintain a short batch window. The strategies exploit the predominately sequential access pattern as well as prior information about the files to be processed per batch job. A simple scheduling approach was developed to execute batch jobs accessing the same files concurrently in order to improve inter-job locality of reference.