**Table 1.**

*Literature summary.*

Tasks are divided into subtask, and to manage the subtask is challenging issue. To handle this challenge, a generalized Nash equilibrium game called parallel scheduling of multiple tasks is developed. For scheduling the task, a distributed task scheduling algorithm was developed via Gauss Spidel-type method [43].

Non-orthogonal multiple accesses-based fog computing framework for industrial IoT system is proposed. Here the task offloading is based on NOMA to the helper node to minimize the delay and energy consumption [44].

A container-based task scheduling algorithm for delay-sensitive and highconcurrency characteristic of fog computing is proposed. The tasks execution is divided into two steps: first to determine whether to accept or reject; second if accepted, then where to forward the task on fog node or cloud. For resource reallocation, a reallocation mechanism is proposed [45].

Tasks with different deadlines are considered. The main objective is to minimize failure probability to meet the different delay deadline. Two queues are considered, low and high-priority queues. For scheduling in the queues, Lyapunov drift plus penalty function is used [46].

To handle the sensitivity of task delay, the laxity-based priority algorithm is suggested. This algorithm is used to decide the priority of the task based on the deadline. Again to minimize energy consumption, an Optimization algorithm based on ant Colony is proposed [47].

The proposed method is based on HH algorithm; it generally focuses on workflow scheduling. The proposed algorithm shows that it reduces the energy consumption and execution time of the task [48].

Delay-sensitive task is considered. DMTO is proposed to identify the optimal subtask size and the TN transmission power [49].

Four criteria are considered in the proposed algorithm: energy dynamic, threshold, waiting time of the task, and communication delay. These criteria are divided into two groups, and based on that, two scheduling and load balancing procedures are performed [50].

Online task scheduling problem in fog computing is discussed. The main focus is to minimize the task slowdown. Deep reinforcement learning and pointer network architecture are combined to propose neural task scheduling [51].

Author in [1] basically focuses on how to reduce the cost. The proposed algorithm efficiently prioritizes the task according to their delay or tolerance level result in higher throughput, which leads to reduce in overall response time and cost (**Table 1**).

#### **3. Motivation**

As we already know that fog is a middle layer between cloud and user. The user's requirement is always QoS. QoS depends on the parameters such as bandwidth, energy consumption, latency, throughput, and cost. So basically fog has to fulfill these requirements of users. Again Fog has limitation such as limited resources and capabilities, but it has an advantage of being nearer to the end devices, which makes it powerful in many aspects such as less latency, less power consumption, and proper utilization of bandwidth. Decision-making on task scheduling is a trending research area. How accurately you can predict the best algorithm on the basis of user's requirement is a challenging issue in fog. Machine learning is making very much progress in this domain. This thing motivates us to use Machine learning algorithm for task scheduling in cloud.
