Parallel Cache Management of a RDBMS 1

Increasing requirements such as higher transaction rates, shorter response times etc., make it necessary to parallelize existing database management systems and to implement them on multiprocessors. Shared memory multiprocessors are not arbitrarily scalable, and thus do not meet future requirements. Therefore, it makes sense to implement databases on large-scale distributed memory multicomputers. This raises the problem how to implement a database cache using shared memory concepts extensively under such a hardware architecture. We describe a solution to this problem by introducing a virtual database cache (VDBC). The VDBC is an algorithmic approach based on the concepts of a virtual shared memory and of explicit synchronization. Our approach is based on the assumption, that interprocessor communication is quite favorable as compared to I/O it is thus interesting to exchange pages with other processors instead of with the disks.