Energy-Efficient Joint Task Assignment and Migration in Data Centers: A Deep Reinforcement Learning Approach

IEEE Transactions on Network and Service Management, Journal Year: 2022, Volume and Issue: 20(2), P. 961 - 973

Published: Sept. 27, 2022

Energy-efficient task scheduling in data centers is a critical issue and has drawn wide attention. However, the execution times are mixed hard to estimate real-world center. It been conspicuously neglected by existing work that decisions made at tasks’ arrival likely cause energy waste or idle resources over time. To fill such gaps, this paper, we jointly consider assignment migration for duration tasks devise novel energy-efficient algorithm. Task can improve resource utilization, required when long-running run low-load servers. Specifically: 1) We formulate as large-scale Markov Decision Process (MDP) problem; 2) solve MDP problem, design an efficient Deep Reinforcement Learning (DRL) algorithm make decisions. DRL more practical real scenarios, multiple optimizations proposed achieve online training; 3) Experiments with have shown our outperforms baselines 14% on average terms of consumption while keeping same level Quality Service (QoS).

Language: Английский

---

Optimizing pre-copy live virtual machine migration in cloud computing using machine learning-based prediction model

Computing, Journal Year: 2024, Volume and Issue: 106(9), P. 3031 - 3062

Published: July 8, 2024

Abstract One of the preconditions for efficient cloud computing services is continuous availability to clients. However, there are various reasons temporary service unavailability due routine maintenance, load balancing, cyber-attacks, power management, fault tolerance, emergency incident response, and resource usage. Live Virtual Machine Migration (LVM) an option address by moving virtual machines between hosts without disrupting running services. Pre-copy memory migration a common LVM approach used in systems, but it faces challenges high rate frequently updated pages known as dirty pages. Transferring these during pre-copy prolongs overall time. If large numbers remaining after predefined iteration page transfer, stop-and-copy phase initiated, which significantly increases downtime negatively impacts availability. To mitigate this issue, we introduce prediction-based that optimizes process dynamically halting when predicted falls below threshold. Our proposed machine learning method was rigorously evaluated through experiments conducted on dedicated testbed using KVM/QEMU technology, involving different VM sizes memory-intensive workloads. A comparative analysis against methods default reveals remarkable improvement, with average 64.91% reduction RAM configurations high-write-intensive workloads, along total time approximately 85.81%. These findings underscore practical advantages our reducing disruptions live systems.

Language: Английский

---

An Efficient Virtual Machine Consolidation Algorithm for Cloud Computing

Entropy, Journal Year: 2023, Volume and Issue: 25(2), P. 351 - 351

Published: Feb. 14, 2023

With the rapid development of integration in blockchain and IoT, virtual machine consolidation (VMC) has become a heated topic because it can effectively improve energy efficiency service quality cloud computing blockchain. The current VMC algorithm is not effective enough does regard load (VM) as an analyzed time series. Therefore, we proposed based on forecast to efficiency. First, migration VM selection strategy increment prediction called LIP. Combined with increment, this accuracy selecting from overloaded physical machines (PMs). Then, point sequence SIR. We merged VMs complementary series into same PM, improving stability PM load, thereby reducing level agreement violation (SLAV) number migrations due resource competition PM. Finally, better LIP experimental results show that our

Language: Английский

---

Krill Herd Algorithm for Live Virtual Machines Migration in Cloud Environments

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(5)

Published: Jan. 1, 2023

Green cloud computing is a modern approach that provides pay-per-use information and communication technologies with minimal carbon footprint. Cloud enables users to access resources without the need for local servers or personal devices operate applications. It allows businesses developers infrastructure hardware conveniently. Consequently, this results in growing demand data centers. becomes crucial maintaining economic environmental sustainability as centers use disproportionate energy. This points energy consumption being important topics research computing. paper introduces two-tiered VM placement algorithm. A queuing model proposed at first level handle many requests. Models such simulation are easily implemented validated using model. also an alternate method allocating tasks servers. Next, multi-objective algorithm based on Krill Herd (KH) Basically, it maintains balance between resource utilization.

Language: Английский

---

Dynamic Threshold Setting for VM Migration

Lecture notes in computer science, Journal Year: 2022, Volume and Issue: unknown, P. 31 - 46

Published: Jan. 1, 2022

Language: Английский

---

RLPRAF: Reinforcement Learning-Based Proactive Resource Allocation Framework for Resource Provisioning in Cloud Environment

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 95986 - 96007

Published: Jan. 1, 2024

Recent developments in cloud technology enable one to dynamically deploy heterogeneous resources as and when needed. This dynamic nature of the incoming workload causes fluctuations environment, which is currently addressed using traditional reactive scaling techniques. Simple approaches affect elastic system performance either by over-provisioning significantly increases cost, or under-provisioning, leads starvation. Hence automated resource provisioning becomes an effective method deal with such fluctuations. The aforementioned problems can also be resolved intelligent techniques assigning required while adapting environment. In this paper, a reinforcement learning-based proactive allocation framework (RLPRAF) proposed. simultaneously learns environment distributes resources. proposed work presents paradigm for optimal merging notions automatic computation, linear regression, learning. When tested real-time workloads, RLPRAF surpasses previous auto-scaling algorithms considering CPU usage, response time, throughput. Finally, set tests demonstrate that suggested strategy lowers overall expense 30% SLA violation 77.7%. Furthermore, it converges at optimum timing demonstrates feasible wide range real-world service-based applications.

Language: Английский

---

A Stochastic Hill Climbing Approach for Power Efficiency in Cloud-Based Systems

Published: Nov. 15, 2024

Language: Английский

---

Cloud Computing Virtual Machine Migration Algorithm Based on Deep Reinforcement Learning and Demand Awareness

Published: Aug. 16, 2024

Language: Английский

---

SPM: A Novel Hierarchical Model for Evaluating the Effectiveness of Combined ACDs in a Blockchain-Based Cloud Environment

Applied Sciences, Journal Year: 2022, Volume and Issue: 12(18), P. 9230 - 9230

Published: Sept. 14, 2022

Cloud computing provides blockchain a flexible and cost-effective service by on-demand resource sharing, which also introduces additional security risks. Adaptive Cyber Defense (ACD) solution that continuously changes the attack surface according to cloud environments. The dynamic characteristics of ACDs give defenders tactical advantage against threats. However, when assessing effectiveness ACDs, structure traditional evaluation methods becomes unstable, especially combining multiple ACD techniques. Therefore, there is still lack standard quantitatively evaluate ACDs. In this paper, we conducted thorough with hierarchical model named SPM. proposed made up three layers integrating Stochastic Reward net (SRN), Poisson process, Martingale theory incorporated in Markov chain. SPM two main advantages: (1) it allows explicit quantification straightforward computation; (2) helps obtain metrics interest. Moreover, architecture each layer be used independently adopted method. simulation results show efficient evaluating various synergy effect their combination, thus improve system configuration accordingly.

Language: Английский

---

Advanced scheduling algorithm for multi resource scheduling with minimum time consumption

Research Square (Research Square), Journal Year: 2023, Volume and Issue: unknown

Published: April 7, 2023

Abstract The scheduling of appropriate resources for cloud workloads is a difficult task, as it depends on the quality service needs applications. Due to their limited data storage and energy capabilities, IoT applications demand high-speed transfer low latency. Many devices generate continuously want store quickly efficiently. Dynamic virtual machine (VM) allocation in centers (DCs) taking advantage computing paradigm. Each VM request characterized by four parameters: CPU, RAM, disk, bandwidth. Allocators are designed accept many requests possible, considering power consumption device's network. Resource time two most significant problems computing. To overcome this problem, paper, author has extended CloudSim with multi-resource minimum model that allows more accurate valuation dynamic scheduling. proposes new algorithm advance algorithm(ASA), which provides better solution other algorithms like Ant Colony Optimization (ACO), Particle Swarm (PSO) Artificial Bee Colony(ABC). also tries reduce give task VM.

Language: Английский

---