Virtual job management and scheduling for media services

Abstract

The necessity for distributed parallel multimedia processing capacity in the multimedia-aware cloud derives from the fact that it must provide both data storage and computational resources to the whole number of interconnected users. When an application or a service becomes popular, the underlying system needs to scale up efficiently thus introducing the distributed factor into the overall scheme. Coordination, job scheduling and efficient task management have always been an area of great importance in the design of distributed computing clusters and will continue to be in the 5G networking era, due to the extensive hardware softwarization it introduces. This paper presents some of the limitations that render efficient job and task management problematic in a vastly distributed system such as a Media-Aware cloud, the most important scientific models that have been developed for describing those limitations, along with the dominant set of software implementations that provide a certain degree of solution to these thorny issues. It becomes clear that the design and implementation phases of a distributed virtual resource management module does not need to take place from scratch, since there are several relatively mature software packages which seem to provide the necessary degree of functionality needed in production-ready deployments.