Dependability and the management of large distributed systems

Abstract

This thesis concerns the design, implementation and application of a dependable management information system to aid in the controlling (moni- toring) of large, complex distributed computer systems. Special attention has been given to using a person centred model of an organisation based on the ANSA "Enterprise Projection" and using fault tolerance techniques to provide continued service and recovery in the event of partial sub-system failure. The information system is accessed using "management workbenches" (implemented in Tclfrk) which access managed resources using "probes" (implemented in C++). Existing "legacy systems" are incorporated in the pro- totype using "integration objects" which "wrap" system software, entry rou- tines, configuration files etc. and enact operation upon logical (physical) resources. Application layer fault tolerance and recovery is implemented using type inheritance whereas remote operations are performed using the Arjuna tool kit. The prototype information system was used to "manage" several resources including: workstations (running SunOS, Solaris and HP-UX) , terminals, printers, disk and tape devices as well as software distributions. A mecha- nism for re-configuring multiple resources (such as workstation clusters and dedicated devices) based on "dependable change schedules" is presented and applied to change and fault management.