원문정보
초록
영어
Cloud computing has based on on-demand access to potentially significant amounts of remote datacenter capabilities in distributed systems environment. By propagation data centers in clouds, computers should have a lower energy consumption. Hence, a good prediction about the amount of needed resources is very helpful for correct decision on energy consumption management. In this paper the Fuzzy Logic approach has been used for decision making about the performance of processors to obtain the estimated amounts of energy consumed by them and have been applied Auto Regressive and Neural Network approaches to predict the future workload. In prediction operation, three modes On, Off and Idle are considered for each server. At times of peak demand all servers must be turned on but when consumption is low, an estimated number of servers that must be turned on is obtained, 10% of remaining servers are idle and others are off. In this case a significant amount of energy is saved compared to the case that all servers are clear. The result show that Neural Network approach has more accurate prediction and better performance than Auto Regressive model.
목차
1. Introduction
2. Related Works
3. Concepts
3.1. Fuzzy Logic
3.2. Auto Regressive Model
3.3. Neural Network
4. Problem Definition
4.1. First phase: Energy Estimation Algorithm using Fuzzy Logic technique (EEFL)
4.2. Second Phase: The Proposed Algorithm for Predicting Amount of Required Energy of Processors
5. Proposed Method
5.1. EEFL Configuration
5.2. EPAR Configuration
5.3. EPNN Configuration
6. Evaluation and Simulation
6.1. Grid5000 Platform
6.2. Estimation Phase
6.3. Prediction Phase
6.4. Prediction Results of both Auto Regressive and Neural Network Methods
6.5. Determine Server State using EPNN
7. Conclusion
References