Dependable Management of Untrusted Distributed Systems

Abstract

The conventional approach to the online management of distributed systemsrepresented by such standards as SNMP for network management, and WSDM for systems based on service oriented computing (SOC)-relies on the components of the managed system to cooperate in the management process, by providing the managers with the means to monitor their state and activities, and to control their behavior. Unfortunately, the trust thus placed in the cooperation of the managed components is unwarranted for many types of systems-such as systems based on SOA-making the conventional management of such systems unreliable and insecure. This paper introduces a radically new approach to the management of distributed systems, called governance-based management (GBM), which is based on a middleware that can govern the exchange of messages between system components. GBM has a substantial ability to manage distributed systems, in a reliable and secure manner, even without any trustworthy cooperation of the managed components. And it can fully incorporate the conventional management techniques wherever such cooperation can be trusted. GBM also supports a reflexive mode of management, which manages the management process itself, making it safer. However, GBM is still a work in progress, as it raises several open problems that needs to be addressed before this management technique can be put to practice.1 Reactive management also differs from general software management, which involves, testing, debugging, people management, etc. 2 By "managers" we mean either people, such as operators, or software components designed for various management tasks.trusted to provide managers with means to monitor their state and activities, and to control their behavior. Standards have been developed to facilitate such cooperation by components. The first of these standards was SNMP (Simple Network Management Protocol)[7], developed about 20 years ago for the management of networks. An analogous standard, called WSDM (Web Services Distributed Management) [16], has been devised more recently for the management of systems based on service oriented computing (SOC). Moreover, the reliance on the cooperation of components is common to most, if not all, recent approaches to the reactive management of distributed systems. In particular, autonomic systems [28] have been conceived to be composed of "autonomic components" that are to be designed in conformance with the policies of the system they are part of. Same is true for the following attempts: the concept of Self-Managed Systems by Kramer et al. [15]; the various versions of Policy Based Management (PBM) [8,17]; and several attempts at management via Computational Reflection [9,1]. Some of these works employ either SNMP or WSDM standards; other use different types of cooperation.Unfortunately, the trust thus placed by the conventional management techniques in the cooperation of the managed components is unwarranted for many types of systems. This is the case, in particular, for the application layer of many distr...