In mobile edge computing (MEC) systems, enhancing the learning capabilities of edge nodes through Artificial Intelligence (AI) can improve the efficiency of dynamically allocating resources. For the scenario of edge service migration, various tasks in lightweight IoT devices are offloaded to edge nodes and the services on edge nodes are migrated adaptively to another available node nearer to the users while they move. To speed up network application loading during service migration, this paper proposes an intelligent trace-driven predicting approach (ITPA) that improves the efficiency of I/O scheduling and the hit ratio of caching when migrating services between resource-constrained edge nodes. First, based on the characteristics of sequential access to the binary codes of an application during its startup progress, the request loading list is generated by tracing key I/O requests at that phase. Then, an intelligent algorithm is designed to search and select the key I/O requests in the loading list. Finally, the efficiency of data acquisition is improved by implementing a prefetch strategy for the client side and a three-level caching strategy for the server side. Experimental results show that the ITPA reduces the service startup time during stateless migration.

在移动边缘计算（MEC）系统中，通过人工智能（AI）增强边缘节点的学习能力，可以提高动态分配资源的效率。对于边缘服务迁移的场景，轻量级物联网设备中的各种任务被卸载到边缘节点，边缘节点上的服务在移动的同时自适应地迁移到离用户更近的另一个可用节点。为了加快服务迁移过程中的网络应用程序加载，本文提出了一种智能跟踪驱动预测方法（ITPA），在资源受限的边缘节点之间迁移服务时提高了 I/O 调度效率和缓存命中率。首先，基于应用程序在启动过程中顺序访问二进制代码的特点，请求加载列表是通过跟踪该阶段的关键 I/O 请求生成的。然后，设计了一种智能算法来搜索和选择加载列表中的关键 I/O 请求。最后，通过在客户端实现预取策略和在服务端实现三级缓存策略，提高了数据获取的效率。实验结果表明，ITPA 减少了无状态迁移过程中的服务启动时间。