The metaverse is a pioneering cyber-physical space that seamlessly blends the physical and virtual worlds, demanding a fully immersive and highly interactive user experience. However, existing communication designs and traditional offloading and allocation approaches fall short of meeting the dynamic network conditions and real-time performance demands of the metaverse. To tackle these challenges, we develop a wireless transmission architecture optimized for metaverse computation offloading and resource allocation. This design leverages the Rayleigh fading and Multiple Input Multiple Output (MIMO) technology to optimize transmission paths, enhancing the reliability and efficiency of signal transmission in high-concurrency connections. We further introduce MetaTMCO (MetaTransformer andMAPPO basedComputationOffloading), a dynamic online computation offloading and resource allocation joint optimized method that meets critical QoS requirements for metaverse applications. MetaTMCO uses a metric named Interaction Frequency (IF) to evaluate resource competition between users and resource interaction between users and the environment, combining QoS to maximize utility under resource constraints. By integrating a Transformer-based encoder-decoder within the Multi-Agent Proximal Policy Optimization (MAPPO) algorithm, MetaTMCO mitigates inefficiencies from partial observations, enabling dynamic optimization of real-time fast offloading strategies across multiple agents. Experimental results demonstrate that MetaTMCO significantly outperforms other approaches in metaverse environments, achieving superior strategy optimization and resource efficiency.

We propose a new workflow task execution time prediction approach, DAG-FGL, by integrating Flash attention mechanism with a GraphLSTM model. It addresses the challenge of low task execution time prediction accuracy in the presence of complex dependencies among workflow subtasks. The GraphLSTM model captures and conveys subtask dependencies through the adjacency matrix of sub task relationships modeled as a directed acyclic graph (DAG). The Flash attention mechanism enhances the model by incorporating customized positional encoding of subtask priority. The encoding ensures that the model accurately reflects each subtask's importance and relative order when calculating attention weight. Therefore, DAG- FG L can more accurately predict task execution time in the context of complex dependencies. Experimental results show that DAG-FGL outperforms the best-performing baseline model. It achieves 7.82 % to 44.52 % improvements in prediction accuracies over the best-performing baselines across three cloud workflow datasets of varying lengths.