الفهرس 
الغلافOnly 14 pages are availabe for public view
 |  | from | 124 |  |  |
| | | from | 124 | | |
Abstract
Over the last decade, the dominant virtualization technology was hardware virtualization (Virtual Machines) which works by emulating a full hardware machine hosting a full operating system. Virtual Machines (VMs) are characterized by energy wastefulness. Currently, a more resource efficient model called containers type virtualization has appeared. Containerization has revolutionized datacenters from being an infrastructure-oriented to being application oriented. Containers have both superior performance and higher resource efficiency when compared to hypervisors that’s why containers are the major deployment model in cloud environments. Away from the success of container type virtualization, it is still required to have a huge ecosystem that tackles the level of management complexities faced by cloud service providers.
The proposed work tackles the challenges of elasticity, power optimization, and network resources efficiency. The proposed work is simulated over a custom built Container Simulator. The results showed that the algorithm is capable of elasticizing the infrastructure more efficiently by 60 % and save up to 20% of the infrastructure power consumption in the midutilization scenario. The results also showed that the algorithm efficiently utilizes the network resources. The algorithm is compared against a set of selection polices where it showed that the least full outperformance its peers in the elasticity and power consumption criterial, while the least pull has a superior performance when it comes to the network traffic minimization. Moreover, the algorithm was tested against different testing thresholds. The results showed comparable results in the 75% and 50% thresholds which vindicate the performance advances of minimizing network traffic.
In addition, our work includes a novel container and host selection policies that can be integrated with different container based cloud deployment and algorithms. The results of different metrics demonstrated the existence of a relation between Container and Host selection Policies. Finally, the results revealed the superiority of the Max-Variance and Mostusage as host and container selection policies.