Article Preview
TopIntroduction
Task scheduling is covered under an area of scheduling algorithms that assist in assigning incoming tasks to the most probable processing nodes. To design a task scheduling model, multiple signal processing and machine learning blocks are used. According to the authors Gill and Buyya (2019), these blocks include but are not limited to task grouping (clustering), group information collection, task dispatcher unit, resource grouping, resource mapping, and task execution units. Figure 1 depicts a typical task (job) scheduling model, where data flow between different units can be observed. In this figure, input tasks (jobs) are initially given to a grouping & selection service module, wherein different clusters of tasks are formed.
Figure 1. A typical task scheduling model
These clusters are formed based on context-aware scheduling metrics, which include, task length, deadline, mean waiting time, mutual task dependency, etc.Clusters are ranked according to their priority of execution, and given to the information collection agent. This agent checks available resources and maps them with each task for execution. The final execution is performed by the task dispatcher unit, which maps the tasks with selected resources, and updates resource capacity values. Based on this process, each task is assigned to a given resource, and its execution is completed. A wide range of system models have been developed for mapping tasks with their relevant resources and each one depends on the nature of processing delay, computational complexity, deployment cost, and other task & resource-related parameters. A survey of some models initially enacted is discussed in the next section, which reviews their nuances, and brief algorithmic details. As per this review, it is started to notice that a very small portion of these types of models utilizes service-level agreement (SLA) enforcement, which limits their real-time applicability. This is because SLAs are agreements that assist in fair scheduling from task-level & resource-level perspectives.
Section 3 proposes an improved task-side service level agreement model for efficient task-scheduling via bioinspired deadline-aware pattern analysis based on the study and observation done in the survey section. On a variety of datasets from the Parallel Workload Archive, the proposed model's performance was evaluated. which include workload logs from NASA, Intel, and other real-time engines. Based on this evaluation, the proposed model was observed to perform better than reviewed models in terms of scheduling efficiency, task diversity, and deadline hit ratios, which greatly facilitates the model's application in real-time deployments. Finally, this paperwork concludes by making a few insightful observations about the proposed model and offering suggestions for ways to enhance it even more.
TopLiterature Review
For mapping tasks using cloud resources, a wide range of scheduling models are proposed, and each of these models has a different performance. This performance varies w.r.t. number of tasks being scheduled, the configuration of machines, etc. For instance, the research that is published by authors Hirth et al. (2019), Sarvabhatla et al. (2017), and Hosseinimotlagh et al. (2014) provides a variety of SLA-based models, such as energy-efficient greedy scheduling, cooperative two-tier energy-aware scheduling, and timely completion SLA (ECTSLA). The performance of these algorithms is slower than that of non-SLA models because they aim to improve scheduling efficiency by creating forecasts of activities and mapping them with time- or energy-bound limits. You could improve this performance by referring to the work of authors Ashouraei et al.(2018), and Sharma and Bharti (2014), who specify parallel SLA models (PSLA) and swarm optimization models. This work was done to help raise this performance. These models aim to achieve optimal internal performance measurements, which, in turn, helps to increase the efficiency of cloud task scheduling.