A study of distributed data processing network survivability
Mohammad Mirhakkak · 1981
Distributed Data Processing (DDP) is being used to bring the computer to the job rather than the job to the computer. These systems employ a combination of computers and telecommunication facilities to provide a distribution of computational functions over many computer sites. A key feature of distributed processing systems is their capacity to remain operational when confronted with component failure. The emphasis of this research is to study the survivability of DDP systems. A model of a DDP system is presented in which data resources are shared among the sites. Each site has been assigned certain programs or tasks to execute. For each program to execute, it should access certain data files which are resident at other nodes. Any destruction of components affects this accessibility. A survivability measure is introduced which depends on a survivability vector whose components indicate whether or not a program can be executed. This measure is investigated for both deterministic and probabilistic models of DDP systems. To quantify this survivability measure, a weighting factor is assigned to each program, which is a function of survivability vector, i.e., it depends on the combination of programs that have survived. The measure is formulated with respect to a state vector whose elements represent the state of the components of the DDP system. Further, a hierarchical approach has been devised to compute the value of the survivability measure which reduces the computational time. The survivability measure is used to evaluate the survivability performance of DDP systems, and to make comparisons among DDP systems. In the remainder of this thesis, some computational results are presented which investigate the effect of topology, addition or deletion of communication links, data file addition or deletion, and data movement on the survivability measure.