Behavior Isolation in Enterprise Systems

Abstract

A barrier to creating the platform-independent services envisioned by middleware-based development infrastructures is the level of performance robustness of the distributed applications created with them, in lieu of unpredictable variations in application behavior or in the resources available for satisfying user requests. Our goal is to improve the behavior locality of distributed applications and to prevent performance (mis-)behaviors from spilling across certain boundaries, since such spillage weakens behavior diagnoses and/or weakens or disables the effects of locally applied control or management methods. Toward these ends, we develop a novel software abstraction, termed {em isolation points} (I-points), which can be used to isolate application components or subsystems from each other. The main contributions of this work are Isolation Points, which are software abstractions for monitoring and understanding dynamic runtime behaviors to better isolation application components hence creating more robust distributed applications. Two concrete artifacts using I-points also developed in this thesis are: I(solation)-RMI and I(solatoin)-Queue. I-RMI demonstrates the utility of isolation points in J2EE's RMI-IIOP domain. I(solation)-Queue applies isolation points to message passing systems.