On Multi-Radio Multi-ServerPowered Multi-Access EdgeComputing

Abstract

Highly intelligent, automated and ubiquitous digital world will be hallmark of the coming decade. To achieve this, we need high-speed, highly-reliable connectivity between physical, digital and biological world. In terms of cloud systems, Multi-access Edge Computing (MEC) has been playing a key role in enabling mobile devices to have swift connectivity to resource-rich cloud servers. However, the current state-of-art may be unable to meet the full connectivity and processing demands of the future compute- and bandwidth-hungry applications transpiring the envisioned digital society. To make up for the capacity, 5G and upcoming 6G extend the channel bandwidth. This exacerbates the already daunting spectrum resource scarcity and adds to the cost of the network. To minimize the cost of the network and delay, a solution recently proposed in the literature is the concept of parallel offloading to multiple servers over multiple radios access technologies (RATs) that a mobile device comes equipped with such as Wi-Fi Direct, Wi-Fi and macro-cellular technology such as 5G.

Using multi-radio multi-server powered MEC, we work on minimizing network delay as well as jointly minimizing network and computation delay. To minimize the network delay, we measure the performance on different radios. Using the obtained performance, we optimally utilize the joint capacities of the radios and schedule the traffic in such a way that packet order at the source and destination is maintained thereby completely avoiding packet reordering delay to keep the throughput intact. We develop a Continuous Non-Linear Program (CNLP) that vary the load on the radio access technologies according to their performances. The proposed CNLP is solved through Lagrange?s Multiplier theorem for several constraints. Furthermore, to ensure smooth relay of the MEC traffic, capacity distribution at the relay node is optimized according to the arrival of the MEC traffic. Numerical results show significant improvement in terms of throughput, delay and QoS compared with other techniques using multiple radios for computation offloading.

To jointly minimize network and computation delay, we develop a technique that chooses the most optimal servers. Further, to minimize server migration and to achieve a convergence point in the algorithm, we formulated a max-min based non-linear lexicographic minimization problem. To solve the formulated problem in polynomial time, we transform the non-linear objective function to a linear one and solve it through the simplex algorithm. Based on the obtained network performance and computation delay, we formulate a multi-server multi-radio load distribution problem to optimally utilize the available capacities of the radios. This problem is solved using techniques from algorithmic game theory. Illustrative numerical results show that proposed technique significantly minimizes computational and network delay.