الفهرس 
صفحة العنوانOnly 14 pages are availabe for public view
 |  | from | 143 |  |  |
| | | from | 143 | | |
Abstract
This thesis handles the issue of energy aware scheduling the workflow’s tasks in smart infrastructure. The cloud environment is the smart infrastructure used in this thesis. The energy aware scheduling has been focused on by many researchers in different fields: industrial, biomedical, military, and others. Decreasing the energy consumed in the cloud environment leads to a decrease in the cost of the cloud’s data center operation and the pollution of the environment. As a result, two new energy aware scheduling algorithms are proposed. The proposed algorithms are addressing the problem of scheduling dependent workflow’s tasks on heterogeneous cloud environment systems. Two challenges have been addressed and resolved by the proposed algorithms. The first is to divide the target deadline fairly across workflow’s tasks while keeping the precedence constraint. The second challenge is to assign the tasks to the proper Virtual Machines (VMs) with the best frequency level, which consumes less energy. It is noticed that decreasing the frequency level of VMs’ processors leads to transient errors. Consequently, the second proposed algorithm considers the reliability requirement with the energy aware scheduler of the workflow’s tasks. In this context, the overall energy consumption for workflow’s tasks executing in the cloud environment will be improved.
The first proposed algorithm is the Energy-Aware Scheduling with Target Deadline Constraint named (EASTD) . Two well-known techniques are used to decrease the energy consumption, Dynamic Power Management (DPM), and the Dynamic Voltage Frequency Scaling (DVFS) techniques. The new EASTD algorithm uses the DVFS technique to decrease the voltage and frequency pair of VMs’ processors. The EASTD algorithm is divided into three phase. The first phase distributes the target deadline fairly across the tasks of the workflow. The second phase arranges the workflow’s tasks in ascending order according to the calculated sub-deadline values. Afterward, it smartly tries to assign the workflow’s tasks to the proper virtual machine with the least frequency level. The third phase tries to exploit the spare time that exists in VMs’ lists to decrease the VMs chosen frequency levels without violating the deadline constraint.
The second proposed algorithm is the Smart Energy and Reliability Aware Scheduling named (SERAS) . The new SERAS algorithm also uses the DVFS technique to decrease the voltage and frequency pair of Virtual Machines (VMs) processors. However, it takes into consideration the reliability requirement, where the checkpointing technique with the rollback-recovery is used. In the SERAS algorithm the overhead taken to detect and correct errors is calculated which increases the execution time of tasks. The SERAS algorithm is divided into three phases. The first phase prepares the workflow’s task set to the second phase. The second phase divides the target deadline across the workflow’s tasks. The third phase assigns tasks to the best virtual machine in an intelligent way while executing the task before its sub-deadline value and decreasing the energy consumption.
Eventually, experiments are carried on five different real workflows generated from the scientific toolkit. The workflows used are Montage, CyberShake, Sipht, Epigenomics, and LIGO Inspiral, which are used to consistently indicate the suitability and reliability of the proposed algorithms in the real-world systems.
Keywords: Energy Efficient, DVFS-Enabled, Green Cloud Computing, Task Scheduling Algorithm, Reliability, DAG Workflow, Heterogeneous System