Thor is an object-oriented database system designed for use in a heterogeneous distributed environment. It provides highly-reliable and highly-available persistent storage for objects, and supports safe sharing of these objects by applications written in different programming languages. Safe heterogeneous sharing of long-lived objects requires encapsulation: the system must guarantee that applications interact with objects only by invoking methods. Although safety concerns are important, most object-oriented databases forgo safety to avoid paying the associated performance costs. This paper gives an overview of Thor's design and implementation. We focus on two areas that set Thor apart from other objectoriented databases. First, we discuss safe sharing and techniques for ensuring it; we also discuss ways of improving application performance without sacrificing safety. Second, we describe our approach to cache management at client machines, including a novel adaptive prefetching strat...

In a large software system that is required to be dependable, preventing the spread of damage from one system component to another is important. Damage confinement both allows damage to be assessed in a reasonable way once an error is detected and prevents a fault in a single component from causing sudden collapse of the entire system. This paper examines techniques for constraining the spread of damage, both the kinds of constraints required and the means for enforcing those constraints. Techniques developed primarily or exclusively for uses other than fault tolerance are described and examined for suitability in confining damage. The influence on damage confinement of the level of dependability required, e.g., for safety-critical systems versus other systems, is also discussed. 1 Introduction One of the four phases in the classical decomposition of fault tolerance is damage confinement and assessment [13]. Damage confinement attempts to constrain the spread of errors fro...

Safe sharing is a desirable feature of an object oriented database because it protects valuable database objects from program errors in application code. It is especially desirable in a heterogeneous environment in which applications are written in various programming languages, many of which have unsafe features. However, safe sharing is not without its potential performance costs. This paper explores these costs. It describes a number of techniques that improve performance without sacrificing safety, and presents results of experiments that evaluate their effectiveness. The results show that some of these techniques are very promising, allowing safe sharing to be achieved with essentially no performance penalty.

Abstract not found