System Stability Under Adversarial Injection of Dependent Tasks

Technological changes (NFV, Osmotic Computing, Cyber-physical Systems) are making very important devising techniques to efficiently run a flow of jobs formed by dependent tasks in set servers. These problem can be seen as generalizations the dynamic job-shop scheduling problem, with rich dependency patterns and arrival assumptions. In this work, we consider computational model distributed system servers which jobs, that continuously arriving, have executed. Every job is (i. e., each task may wait for others completed before it started), has executed one The their properties assumed controlled bounded adversary, whose only restriction cannot overload any server. This non-trivial generalization Adversarial Queuing Theory Borodin et al. , and, like model, focuses on stability system: whether number pending at all times. We show multiple results instability adversarial under different combinations policy used servers, rate, dependence between jobs.