Mobile services such as Augmented Reality (AR) and online gaming require high computing and storage resources with real-time data exchange with servers. However, mobile devices are resource-constrained and cloud computing incurs additional network latency which makes them dissatisfactory to use such mobile services. To tackle this challenge, service caching has been emerging as a critical technology that provides services by leasing resources from mobile edge computing (MEC) servers located close to the users. Unlike content caching, which stores data only for delivery, service caching utilizes both computing and storage resources, and thereby should be jointly considered with code offloading which offloads the processing workloads of mobile devices to servers. Therefore, the conventional caching strategy that caches the most popular services may not be the best choice, and the intertwined nature of service caching and code offloading policies makes the system operation complex. In this paper, we formulate an average cost (energy consumption, offloading payment) minimization problem of mobile devices constrained by service queue stability in a three-layer (mobile/MEC/cloud) network architecture. We apply a statistic-based dynamic optimization framework to derive an online-fashioned interTWined Interplay between Storage and Transmission, namely TWIST algorithm that jointly makes decisions of service caching, CPU clock frequency of mobile devices, and code offloading. Next, via theoretical analysis, we show the trade-off relationships between the average cost of mobile devices and service delay. Finally, extensive simulations demonstrate that TWIST can save up to 51% of cost for the same delay of 2.41sec. compared to the existing algorithms.

As deep learning technology becomes advanced, mobile vision applications, such as augmented reality (AR) or autonomous vehicles, are prevalent. The performance of such services highly depends on computing capability of different mobile devices, dynamic service requests, stochastic mobile network environment, and learning models. Existing studies have independently optimized such mobile resource allocation and learning model design with given other side of parameters and computing/network resources. However, they cannot reflect realistic mobile environments since the time-varying wireless channel and service requests are assumed to follow specific distributions. Without these unrealistic assumptions, we propose an algorithm that jointly optimizes learning models and process/network resources adapting to system dynamics, namely, VisionScaling by leveraging the state-of-the-art online convex optimization (OCO) framework. This VisionScaling jointly makes decisions on 1) the learning model and the size of input layer at learning-side and 2) the GPU clock frequency, the transmission rate, and the computation offloading policy at resource-side every time slot. We theoretically show that VisionScaling asymptotically converges to an offline optimal performance with satisfying sublinearity. Moreover, we demonstrate that VisionScaling saves at least 24% of dynamic regret which captures energy consumption and processed frames per second (PFPS) under mean average precision (mAP) constraint via real trace-driven simulations. Finally, we show that VisionScaling attains 30.8% energy saving and improves 39.7% PFPS while satisfying the target mAP on the testbed with Nvidia Jetson TX2 and an edge server equipped with high-end GPU.

Service caching has been emerging as a key technology that can overcome the hardware limitations of mobile devices by leasing resources from mobile edge computing (MEC) servers. This technology is inherently different from content caching where caching performance can be maximized by caching the most popular contents since most of the services require workload processing from CPU/GPU in servers. Thereby caching the most popular services on the MEC server may not be the best choice. Besides, when the service caching system is designed together with code offloading policy, the problem becomes more difficult because two control parameters are tightly intertwined with each other. In this paper, we formulate an average cost minimization problem of mobile devices constrained by service queue stability in a three-layer (mobile/MEC/cloud) network architecture. We apply a statistic-based dynamic optimization framework to derive an online-fashioned interTWined Interplay between Storage and Transmission, namely TWIST algorithm that jointly makes decisions of service caching, CPU/GPU clock frequency of mobile devices, and code offloading. Next, via theoretical analysis, we show the trade-off relationships between average cost of mobile devices and service delay. Finally, extensive simulations demonstrate that TWIST can save up to 51% of cost for the same delay 2.41sec. compared to the existing algorithms.