**2. Related Work**

In this section, we review the existing related works to clarify the motivation of the proposed algorithm. Over the past several years, MEC researchers have focused on maximizing computational efficiency and minimizing energy consumption of IoT devices [11–16]. In [11], the authors proposed joint offloading and computation energy efficiency maximization algorithm for MEC system. They proposed novel computation efficiency indicator and solved the problem by using iterative and gradient descent method. However, this algorithm did not consider energy harvesting. In [12], energy efficient task offloading algorithms for non-orthogonal multiple access (NOMA) MEC environment is presented. This algorithm determined the uplink power control solution, and then solved the task offloading partition and time allocation. However, this scheme also did not consider energy harvesting. The authors of [13] proposed the delay constraint offloading algorithm. This algorithm solved the problem of minimizing energy consumption, assuming that MEC server can charge IoT devices. In [14], the authors proposed a bound improving branch and bound approach to minimize energy consumption of IoT device in which orthogonal frequency division multiple access (OFDMA) was considered for uplink transmission. They focused on the energy consumption of the IoT device, under the consideration of computation offloading, subcarrier allocation, and computing resource allocation. In [15], an IoT device offloading scheduling algorithm in wireless power transfer environment is proposed. In this algorithm, the IoT device decides whether to compute the task itself or offload the task to MEC server. A cooperative partial computation offloading algorithm is proposed for MEC and cloud server environment in [16]. The authors proposed the branch and bound approach to solve the single MEC scenario, and then expanded it to multiple MEC and cloud scenario. For the multiple scenarios, an iterative heuristic MEC resource allocation algorithm was proposed. The authors of [9] proposed an algorithm for computation offloading scheduling of MEC server. They minimized the energy consumption of the MEC server as well as guaranteed the stability of the task buffer. However, it is difficult to apply this algorithm in EH environment since they did not consider energy harvesting. These previous studies [11–16] focused on the energy efficiency of IoT devices. However, this paper focuses on the energy efficiency scheduling of MEC server, i.e., MEC server has to provide stable offloading service. Due to the advancement in energy harvesting techniques, research about IoT

systems with energy harvesting attracts significant attention [10,17–19]. In [17], reinforcement learning based offloading algorithm is proposed. In this algorithm, energy harvesting IoT devices decide their offloading rates according to battery levels. However, the target of this algorithm is not suitable to the proposed environment where MEC is harvesting energy. In [18], the authors proposed dynamic computation offloading algorithm with a special focus on computation capacity of MEC server. In this algorithm, the energy harvesting is not primarily considered in the offloading scheduling; it is only mentioned that it can extend network life time. A healthcare IoT system was proposed in energy harvesting environment by the authors of [19]. This algorithm is also not suitable for the EH MEC environments since it aims to protect user privacy. The authors of [10] proposed a Lyapunov optimization based algorithm for energy harvesting IoT devices. They assumed that the IoT device is equipped with an energy harvesting module and harvests electricity from the module. Each IoT devices schedules its computation frequency based on the energy harvesting status. All of the works in [10,17–19] assumed that IoT devices were equipped with energy harvesting modules and tried to solve the offloading scheduling problem under battery constraints. However, these algorithms are not suitable for the proposed EH MEC environment where the MEC server is equipped with energy harvesting module, i.e., our proposed scheme aims at the stable operation of MEC server in energy harvesting environment, but previous studies aimed at survivability of IoT devices. In [20], the authors proposed MEC scheduling algorithm in EH MEC environment. They modeled the MEC battery and harvested energy, and minimized the service delay and operation cost via reinforcement learning. The energy harvesting model is similar to the model of our proposed algorithm. However, they did not consider the deadline constraint in their problem formulation. In this paper, we consider the constraints of service deadline.
