Energy-Efficient Offloading for DNN-based Smart IoT Systems in Cloud-Edge Environments
ABSTARCT :
Deep Neural Networks (DNNs) have become an essential and important supporting technology for smart Internet-of-Things (IoT) systems. Due to the high computational costs of large-scale DNNs, it might be infeasible to directly deploy them in energy-constrained IoT devices. Through offloading computation-intensive tasks to the cloud or edges, the computation offloading technology offers a feasible solution to execute DNNs. However, energy-efficient offloading for DNN based smart IoT systems with deadline constraints in the cloud-edge environments is still an open challenge. To address this challenge, we first design a new system energy consumption model, which takes into account the runtime, switching, and computing energy consumption of all participating servers (from both the cloud and edge) and IoT devices. Next, a novel energy-efficient offloading strategy based on a Self-adaptive Particle Swarm Optimization algorithm using the Genetic Algorithm operators (SPSO-GA) is proposed. This new strategy can efficiently make offloading decisions for DNN layers with layer partition operations, which can lessen the encoding dimension and improve the execution time of SPSO-GA. Simulation results demonstrate that the proposed strategy can significantly reduce energy consumption compared to other classic methods.
EXISTING SYSTEM :
? A constant computing intensity (in computing cycles per unit data) exists for such tasks, and we can use it to capture the effective computing capability of a specific device.
? Existing literature, such as, has leveraged this observation to characterize deep learning workloads, and in this work, we adopt it to estimate the computing cycles amount given the partitions and DNN layers.
? Some other works target to optimize the utilization of existing hardware.
? We implement a multi-device prototype using heterogeneous edge devices, and evaluate CoEdge on four widely-adopted DNN models to corroborate its superior performance.
DISADVANTAGE :
? Major problems while designing M-IoT networks architecture, protocols, and computing schemes are explored to provide stable IoT architecture.
? The problems of reliability-aware optimal computing offloading and resources allocation are considered in.
? They present the computational offloading and allocation problem as a combinatorial optimization problem while considering offloading valuable basic (OVB) constraints.
? Based on the computation rate, a maximization problem is studied in for the decrease in propagation loss that severely affects the harvested energy and computation performance of UE.
PROPOSED SYSTEM :
• In this paper, we propose CoEdge, a distributed DNN computing system that orchestrates cooperative DNN inference over heterogeneous edge devices.
• CoEdge utilizes available computation and communication resources at the edge and dynamically partitions the DNN inference workload adaptive to devices’ computing capabilities and network conditions.
• While DeepThings leverages a layer fusion technique to reduce communication overhead, CoEdge proposes to optimize workload allocation to maximally utilize heterogeneous edge resources.
• We propose CoEdge, a distributed DNN computing system that orchestrates cooperative inference over heterogeneous devices to minimize system energy consumption while promising response latency requirement.
ADVANTAGE :
? The enhancement in the computational performance of active UEs by the user cooperation technique is investigated in, where the inactive UEs use their harvested energy for the help of active ones.
? We can observe better performance of CEDOT because of the proposed comprehensive cost function.
? Therefore, the chances of selecting the offloading policy with minimum cost decreases and the performance of TOT and ROT decreases.
? However, the performance of the proposed work is higher than all the other techniques and is almost equal to EEDOT with the advantage of low energy consumption and time delay for computational offloading and partitioning.
? In this paper, we proposed a comprehensive cost function for energy efficient computational offloading in MEC.
|