Multi-Granularity Decomposition for Componentized Multimedia Applications based on Graph Clustering

With the number of smart terminals increasing sharply, the emerging smart network applications, like virtual/augmenting reality games, interactive video streaming, appear explosive growth, draining not only communication but also computing and storage resources from the links and computing nodes near the network edges. Therefore, the high-load mobile network edge becomes the bottleneck curbing the network service provision capabilities. One promising way to deal with this issue is to utilize mobile terminal resources by offloading the caching, transmission, or computing tasks to other available terminals nearby. Recent technology developments, such as device-to-device (D2D) direction communication, computing dense chips, are making terminals more powerful to enable the solution. However, current network applications (or tasks, in the computing context) are getting too large to be fully served by a single network terminal. But, luckily, future network applications are developing towards a componentized architecture to enable beneficial features, like rapid iteration, high flexibility. Motivated by these factors, we propose to decompose a network application into small subtasks to enable cooperation among multiple nodes, so that terminals can take part in complex tasks. Thus, resources in network terminals can be fully utilized, relieving the load pressure at the network edge. The paper then focuses on the multi-granularity decomposition method for componentized network applications which produces a series of sets of decomposed subtasks in multiple granularities to meet diversified resource provision conditions at network edge scenarios, while the expected overall execution cost is reduced.