Functional Active Monitoring System FLAME

Abstract

The problem of operative monitoring of the state of a complex computer system aimed at preparing information required for operative control of the system in critical situations is a key problem on the stage of the system operation. An operative response to emergency situations in complex network structures and high-performance mass-parallel computer systems by means of traditional state control mechanisms, without help of modern technologies and without automation of this process, is now impossible. Nowadays, systems created for active monitoring of large-scale network structures, such as HP OpenView, Sun Net Manager, and others, are applied to systems with mass parallelism and to metacomputer structures. Such an approach has a number of disadvantages due to the fact that specific features of these systems are not completely taken into account and because of limited possibilities to make alterations during the system development. This paper presents results of the work on the creation of an active monitoring system that takes into account specific features of high-performance, multiprocessor, mass-parallel clusters. The paper discusses specific features of the system architecture, dynamic database structure, configuration description language, its syntax and semantics, and special forms. Examples illustrating the system functionality as applied to the supercomputer system MVS-1000M are presented.