Scheduling Algorithms in Cloud over different scenarios
Optimizing resource utilization in real-world systems has become crucial in recent years. Traditional scheduling algorithms have been widely used for decades, such as First-Come-First-Serve (FCFS), Shortest Job First (SJF), and Round Robin (RR). However, these algorithms are only partially optimized for large-scale problems where exact numerical methods are impractical due to their computational complexity. As a result, researchers have been exploring more optimized algorithms, such as Particle Swarm Optimization (PSO), which has shown remarkable results. In this blog, I will discuss a research paper that compares the performance of PSO as a task-scheduling algorithm to other scheduling methods in different scenarios.
The research paper evaluated the effectiveness of PSO and compared it to other scheduling methods, including FCFS, SJF, and Round Robin. The study’s primary objective was to determine whether PSO’s heuristic optimization approach can effectively manage large-scale problems where exact numerical methods are often impractical due to their computational complexity. To provide a more comprehensive analysis, the researchers also included FCFS, SJF, and Round Robin algorithms, which have been widely used in the field for a long time. The study examined the impact of different task scheduling algorithms on the makespan time of various real-world systems.
The research was conducted by a team of interns, including Ishaan Verma, Aditya Ojha, and under Professor Dr. Vijay K Madisetti at the Georgia Institute of Technology, Atlanta, U.S.A. The study utilized Cloudsim, a cloud environment simulator, to conduct simulations using a dataset with 1491 entries and six features to compare the performance of the different algorithms. The dataset included four-run results to ensure that the variation in values is limited. The performance of different machine types, High Performance (HP) and Low Performance (LP), in two real-world scenarios, were also evaluated. These scenarios included a Short Operation (SO) task, which occurs in small bursts, and a Long Operation (LO) task, which is sustained more extended.
The study encompassed the examination of several key algorithms, including FCFS, SJF, Round Robin, and PSO, and their respective operations of LPLO, HPLO, HPSO, and LPSO. The researchers leveraged several Python libraries to process large-scale data and plot the data in graphs to identify and analyze trends.
The study results showed that PSO was more effective in managing large-scale problems than other scheduling methods, including FCFS, SJF, and Round Robin. The PSO algorithm also improves real-world systems’ efficiency, which can lead to significant cost savings and increased productivity.
The study conducted an extensive series of simulations to evaluate the efficacy of algorithms for task scheduling. The simulations involved incremental increases of 10 tasks, ranging from a baseline of 100 tasks to a maximum of 1500 tasks. From the data generated and graphs plotted, the study found that PSO consistently outperformed other algorithms, such as FCFS, SJF, and RR, in all cases. PSO’s heuristic optimization approach may be more effective than traditional scheduling algorithms for real-world systems that involve a large number of tasks.
The study also found that the performance gap between PSO and other algorithms, such as FCFS, SJF, and RR, increases as the number of tasks increases. FCFS was found to be the most unpredictable, with the execution time varying significantly depending on the number of tasks. RR, designed to be a fair allocation algorithm, was found to be very consistent, and SJF scheduling algorithm was found to be less consistent as some processes may experience longer response times than others.
Additionally, the study investigated the performance of low power machines on both long and short operations and found that there was no significant trend observed. This finding suggested that low power machines may not be suitable for managing varying workloads. To better simulate real-world scenarios, the study utilized a high power machine and introduced large variance to the workload, achieving more accurate and meaningful results that can help inform the optimization of task scheduling in cloud-based systems or other systems with varying workloads. However, the use of high power machines may not always be feasible due to cost and resource constraints, and thus alternative methods should be considered when conducting similar investigations.
