اسم الباحث :
Zaid Ali Ashabi Ibrahim Ahmed Al-Baltah
Abstract Resource management is one of the most important methods used to manage distributed systems. The amount of information today is continuously increasing. It is increasingly difficult to store, modify and analyze the large size, variety and speed of data, which complicates the management of resources and the unreliability of the system. There is a shortcoming with several algorithms, the most important of which is reinforcement learning (RL), since it does not handle the scheduling process in an excellent way, causing a slowdown in resource management, which leads to poor reliability. The proposed model improves the work of RL in order to better manage resources. We used the approach of integrating algorithms to address the problem. We used the auto encoder algorithm for deep learning and integrate it with the reinforcement learning algorithm, which helps us improve scheduling performance in order to optimize the management of resources by learning the powerful features of the auto encoder. Keywords: Resource management, Distributed systems, Reinforcement learning, Auto-encoder, Reliability. The important progress of the distributed systems and the nature of its various hardware and software tools, an expansion that meets the need to provide the large information and dynamic services on the Web, which is expanding day by day [1]. The diversity of resources and the complexity of modern communication technologies in distributed systems cause problems in resource management, so distributed systems need to manage their resources in a way that helps them to deal with data more flexibly and reliably for manage resources well [2]. With accumulative range of on-line services or automotive end-to-end solutions and growing range of user’s mistreatment these services, necessities for extremely reliable systems became in dispensable [1]. The distribution of tasks to the resources equally and monitor their performance so that we do not carry any burdens of tasks alone, and this is due to the good management of resources, where all the resources required to carry out its tasks without carrying overload and decide to move tasks from one to another according to the need of tasks, which gives us the availability and reliability [2]. Reliability depends on allocate resources (scheduling) to work well on the network and to keep working on the network even in difficult and volatile conditions, and the processing does not affect the user's requirements and is subject to failure and high availability or performance under the unemployed and this is called reliability [3]. The choices don't seem to be solely taking charge of matching and programming the process capacities of resources and user necessities, they have to traumatize a good form of resource behaviors and performance fluctuations. Services in large-scale distributed environments should be improved [4]. Square measure is needed to remain and continue operative even within the presence of malicious and unpredictable circumstances; that's, their process capability should not be considerably full of the user necessities [5, 6]. In recent years, deep learning has achieved good success in several fields, like machine vision and language process. Compared to ancient machine learning techniques, deep learning contains a sturdy brain power and may build higher use of datasets for feature extraction [7]. As a result of its utility, deep learning becomes extra and additional well-liked and adaptation artificial learning for several researchers to try and do their works. If we can rebuild data using fewer features, these features are a good abstract representation of data. This is how the Auto encoder works. Generally, auto encoder consists of three layers: the input layer, output layer, and hidden layer [8]. Also, the number of nodes in the hidden layer must be lower than the input and output layers. The main objective of the hidden layer is to reduce the error ratio between input and output. Where the education process aims to minimize the incidence of this error. As a result, the hidden layer is able to reproduce the entered data again, so that this hidden layer represents the data attributes. Achieving a reliable resources management is still a challenge in distributed systems for creating resources allocation extremely reliable. However, this paper aims to propose a model IRL. This model is a combination of RL and Auto encoder algorithm, and the objective of this model is to improve the resource management process to achieve reliability. During this analysis we tend to propose hybrid model IRL for features learning in distributed systems, for mapping the tasks to resources. Our approach makes resource management reliable.
سنة النشر :
2017
ملخص البحث :
