Deep neural network (DNN)-task enabled mobile edge computing (MEC) is gaining ubiquity due to outstanding performance of artificial intelligence. By virtue of characteristics of DNN, this paper develops a joint design...
详细信息
Deep neural network (DNN)-task enabled mobile edge computing (MEC) is gaining ubiquity due to outstanding performance of artificial intelligence. By virtue of characteristics of DNN, this paper develops a joint design of task partitioning and offloading for a DNN-task enabled MEC network that consists of a single server and multiple mobile devices (MDs), where the server and each MD employ the well-trained DNNs for task computation. The main contributions of this paper are as follows: First, we propose a layer-level computation partitioning strategy for DNN to partition each MD's task into the subtasks that are either locally computed at the MD or offloaded to the server. Second, we develop a delay prediction model for DNN to characterize the computation delay of each subtask at the MD and the server. Third, we design a slot model and a dynamic pricing strategy for the server to efficiently schedule the offloaded subtasks. Fourth, we jointly optimize the design of task partitioning and offloading to minimize each MD's cost that includes the computation delay, the energy consumption, and the price paid to the server. In particular, we propose two distributed algorithms based on the aggregative game theory to solve the optimization problem. Finally, numerical results demonstrate that the proposed scheme is scalable to different types of DNNs and shows the superiority over the baseline schemes in terms of processing delay and energy consumption.
The surging Deep Neural Network (DNN) based applications are becoming increasingly popular in mobile computing. However, they impose significant challenges for mobile computing, as DNN tasks lead to much more computat...
详细信息
ISBN:
(纸本)9781728109626
The surging Deep Neural Network (DNN) based applications are becoming increasingly popular in mobile computing. However, they impose significant challenges for mobile computing, as DNN tasks lead to much more computation complexity and data volume compared with traditional tasks. To alleviate this, mobile edge computing (MEC) provides a feasible approach through task partitioning and offloading. In this paper, we investigate a DNN based MEC scheme considering multiple mobile devices and one MEC server. To facilitate taskpartitioning, we first develop a processing delay prediction mechanism for typical DNN tasks. To achieve the minimal processing delay as well as to release the computing burden of mobile devices, a mixed integer linear programming (MILP) based DNN task partitioning and offloading mechanism is presented. Evaluations show that our mechanism can achieve up to 90.5% and 69.5% processing delay reduction compared with MEC server only and mobile device only schemes respectively.
Real-time visual computing applications running Deep Neural Networks (DNN) are becoming popular for mission-critical use cases such as, disaster response, tactical scenarios, and medical triage that require establishi...
详细信息
ISBN:
(纸本)9798350331653
Real-time visual computing applications running Deep Neural Networks (DNN) are becoming popular for mission-critical use cases such as, disaster response, tactical scenarios, and medical triage that require establishing ad-hoc edge environments. However, strict latency deadlines of such applications require real-time processing of pre-trained DNN layers (i.e., DNN inference) involving image/video data which is highly challenging to achieve under such resource-constrained edge environments. In this paper, we address the trade-off between end-to-end latency of DNN inference and IoT devices' energy consumption by proposing 'EFFECT-DNN', an energy efficient edge computing framework. The EFFECT-DNN framework aims to strike such balance by employing a collaborative DNN partitioning and taskoffloading strategy. Such strategy also involves resource allocation from IoT devices and edge servers to satisfy DNN inference deadline requirement even when the network bandwidth is on the lower end, which is often the case for critical use cases. The underlying optimization is formulated as a dynamic Mixed-Integer Nonlinear Programming (MINLP) problem is decoupled and solved by convex optimization and a game-like heuristic algorithm. We evaluate the performance of EFFECT-DNN framework on a hardware testbed and using extensive simulations with real-world DNNs. The results demonstrate that the proposed framework can ensure DNN inference deadline satisfaction with significant (similar to 20-30%) device energy savings.
暂无评论