Distributed optimization Grid resource discovery

Mohammad Hasanzadeh, Mohammad Reza Meybodi
2014 Journal of Supercomputing  
Grid computing is a framework for large-scale resource sharing and indexing that evolves with the goal of resource provisioning. In this paper, we develop a distributed learning automata (DLA) based on multi-swarm discrete particle swarm optimization (PSO) approach for Grid resource discovery, called distributed optimization grid (DOG) resource discovery algorithm. This algorithm makes use of swarms of particles for different computational resource metrics while a group of DLA is the control
more » ... t of each swarm of particles. The algorithm takes advantage of the PSO solution diversity to optimize the quality of delivered resource. Moreover, the recommended algorithm uses DLA as a fully distributed model for imitating the Grid infrastructure topology. Our experimental results show that DOG is fast as well as efficient and accurate. rules, where each of them forms an independent virtual organization (VO) [1] . Each VO accompanies with a Grid information service (GIS) [2] [3] [4] leveraging the resource management issues. For instance, an implementation of such a service is the monitoring and discovery system (MDS) [1] of the Globus Grid Toolkit [5] . Highly dynamic networks such as Grid must have a functionality to manage and monitor resource allocation and condition processes. The Grid is responsible for processing the incoming tasks. It utilizes a scheduler to fulfil these tasks. Also, The resource discovery is the process of finding the requested resources of tasks with respect to one or more user criterion performance, price, etc. Furthermore, the resource discovery overlaps with Grid scheduler in terms of resource scavenging and reserving. Ad Hoc Grid illustrates a highly dynamic computing environment. In [6], a resource selection scheme for Ad Hoc Grid is proposed. Coordinate load management enables Grid to efficiently manage computing resources. In [7], a coordinated load management approach is applied to Grid resources to perform resource brokering. Preventing the communication bottleneck is one of the major challenges of Grid. In [8], a direction aware approach for resource discovery in Grid and Cloud is introduced. Resource allocation is a complex problem in Grid and Cloud environments. A task-oriented resource allocation model proposed for Cloud environments in [9] . Resource allocation mechanisms aim for satisfying the user preferences and Grid policies. A Multi-agent based negotiation strategy is proposed in [10] for studying the interactions between Grid providers and consumers. Machine Learning (ML) is a field of Artificial Intelligence (AI), aiming to construct a model that can learn from data. Focusing on the ML algorithms, one can view Reinforcement Learning (RL). RL studies the learning behavior of Agents while interacting with an unknown environment. Learning Automata (LA) [11] is one of the RL's tools which embodies an autonomous unit placed in an unknown environment trying to interact with the environment by taking a series of actions and earning a series of reward or penalty signals. Moreover, DLA [12] is a newfound network of LA that cooperatively collaborates to solve a particular problem. Computational Intelligence (CI) is a multidisciplinary topic in AI which utilizes different computational methodologies to address the complicated real-world problems. Evolutionary Computation (EC) is a subfield of CI that is inspired from evolution biology of species. Moreover, PSO [13] is an EC algorithm that is inspired from social behavior of flocks of birds and schools of fish. LA enable the versatility of AI with a simple integration of probabilistic control parameters into the adaptive control systems. Target monitoring and lifetime scheduling are among challenging issues of wireless sensor networks (WSNs). Moreover, LA models are employed in [14] for solving target coverage problem of WSNs. Classification algorithms try to determine the observations labels based on the set of training data. A Cellular Automata (CA) model is proposed in [15] for binary classification problem. Also, the mobile Ad Hoc networks (MANETs) are decentralized networks constituting of independent groups of nodes. In [16] , LA theory is used for designing a routing protocol for MANET. Furthermore, in [17], a cellular learning automata (CLA) model is introduced for simulating the investment behavior of stock market dealers. Finally, designing a resource discovery protocol for Grid environment is a challenging
doi:10.1007/s11227-014-1289-4 fatcat:evkwukzemnhlbnkgdsyf34ynyu