To tackle the difficulties of running sophisticated deep neural network (DNN) models on mobile devices, split computing presents a viable solution by offloading computations to the edge server. Current split computing schemes typically aim to lower either inference latency or energy use separately; however, optimizing both simultaneously is quite challenging due to numerous shifting factors, such as intensive continuous DNN model inferences, DNN model traits, and device/network conditions. Moreover, in practical applications, edge server overload might lead to substantial queuing delays, adding complexity to the optimization process. This paper outlines a joint optimization problem that simultaneously seeks to minimize both inference latency and energy consumption, with a distinct inclusion of queue clearance latency for an accurate analysis of the continuously generated DNN model inferences. To address this intricate optimization challenge, we introduce a low-complexity heuristic algorithm that sets split point decisions based on the residual energy of mobile devices for each DNN inference cycle. Upon evaluation, our proposed algorithm demonstrates notable improvements by reducing inference latency by between$73.37\%$and$99.39\%$, and cutting down energy usage by between$39.97\%$and$94.67\%$compared to fully local processing on mobile devices.

Distributed serverless edge clouds and split computing are promising technologies to reduce the inference latency of large-scale deep neural networks (DNNs). In this article, we propose a dynamic split computing framework (DSCF) in distributed serverless edge clouds. In DSCF, the edge cloud orchestrator dynamically determines 1) splitting point and 2) warm status maintenance of container instances (i.e., whether or not to maintain each container instance in a warm status). For optimal decisions, we formulate a constrained Markov decision process (CMDP) problem to minimize the inference latency while maintaining the average resource consumption of distributed edge clouds below a certain level. The optimal stochastic policy can be obtained by converting the CMDP model into a linear programming (LP) model. The evaluation results demonstrate that DSCF can achieve less than half the inference latency compared to the local computing scheme while maintaining sufficient low resource consumption of distributed edge clouds.

Training on time-series data generated from mobile networks is a resource-intensive and time-consuming task that encounters various training failures. To cope with this issue, we propose CheckBullet, a lightweight checkpoint system to minimize storage requirements and enable fast recovery in mobile networks. First, CheckBullet determines a checkpointing interval based on the characteristics of the model and the timing of failure occurrences. This approach ensures fast recovery while preserving the existing training runtime. Second, CheckBullet quantizes the weight tensor and eliminates duplicate weights, which significantly reduces the overall checkpoint size, leading to a substantial decrease in storage requirements. Third, CheckBullet selects the minimum training loss among the deduplicated checkpoints and merges the selected checkpoints. This approach reduces recovery time while preserving existing training loss. The experimental results show that CheckBullet can reduce the recovery time by$6\times$to$11\times$barely increasing the training runtime. Furthermore, CheckBullet can save storage requirements by up to 70% while maintaining the minimum training loss.