Cooperative Task Offloading Strategy for Utility Optimization in Blockchain-Enabled IoT Networks

In the convergence of the Internet of Things (IoT) and blockchain, edge computing enables resource-constrained end devices to offload compute-intensive mining tasks to edge servers to enhance their performance or profits. This calls for a task offloading strategy that accounts for the inherent complexity and variability of the environment, while effectively solving a typically NP-hard offloading problem. Traditional one-shot opti-mization methods often struggle to adapt to dynamic conditions. Meanwhile, existing learning-based approaches usually rely on centralized frameworks or independent agents, which are inad-equate for distributed IoT networks. To this end, a cooperative task offloading strategy is proposed for a blockchain-enabled IoT network with multiple edge service providers. Specifically, the offloading problem is first incorporated into a Markov decision process that considers time-varying channel conditions. A multi-agent deep reinforcement learning algorithm with a gradient estimator is then utilized to optimize the long-term mining utility. Following a centralized training and decentralized execution paradigm, this algorithm allows IoT devices to learn collaborative policies during training while making autonomous decisions based on local observations during execution. Experimental results show that the proposed strategy outperforms existing benchmarks in mining utility under varying conditions.