An efficient object manager, a middle layer on top of a storage system, is essential to ensure acceptable performance of object-oriented database systems, since a traditional record-based storage system is too simple to provide object abstraction. In addition, an object storage system–object managers in combination with storage systems - should be extensible to meet the various requirements of emerging applications. In this research, we design and implement an extensible object storage system, called Soprano, in an object-oriented fashion which has shown great potential in extensibility and code reusability. Soprano provides a uniform object abstraction and gives us the convenience of persistent programming through many useful persistent classes. Also, Soprano supports efficient object management and pointer swizzling for fast object access. This paper investigates several aspects of the design and implementation of the extensible object storage system. Our experience shows the feasibility of using an object-oriented design and implementation in building an object storage system that should have both extensibility and high performance.

When accessing objects in an object-oriented database, performance can be greatly improved by prefetching objects efficiently. In this paper we present a new object prefetch policy, dynamic selective eager object fetch (SEOF), which prefetches objects only from selected candidate pages without using any high-level object semantics. Our policy considers both correlations and frequencies of fetching objects when selecting prefetch candidates. Unlike existing prefetch policies, dynamic SEOF utilizes the memory and the swap space of clients efficiently without resource exhaustion. Furthermore, the proposed policy has good adaptability to both the effectiveness of clustering and database size. It also adjusts the degree of prefetching dynamically according to the working environment. We show the performance of the proposed policy through experiments over various multi-client system configurations.

The performance of object access can be drastically improved by an efficient object prefetch policy. In this paper, we present a new prefetch policy, called Selective Eager Object Fetch (SEOF), which prefetches objects only from selected candidate pages without using any high-level object semantics. Our policy considers both the correlations and the frequencies of fetching objects. Unlike existing prefetch policies, this policy utilizes the memory and the swap space of clients efficiently without resource exhaustion. Furthermore, the proposed policy has good adaptability to both the effectiveness of clustering and the database size. We show the performance of the proposed policy through experiments over various multi-client system configurations.