Hide/Show Apps

Optimal dynamic resource allocation for heterogenous cloud data centers

Ekici, Nazım Umut
Today's data centers are mostly cloud-based with virtualized servers to provide on-demand scalability and flexibility of the available resources such as CPU, memory, data storage and network bandwidth. Heterogeneous cloud data centers (CDCs) offer hardware accelerators in addition to these standard cloud server resources. A cloud data center provider may provide Infrastructure as a Service and Platform as a Service (IPaaS), where the user gets a virtual machine (VM) with processing, memory, storage and networking resources, which can be installed with any desired operating system and software. Differently, Software as a Service (SaaS), only enables user access to provided application for example via a web browser without any control of the underlying infrastructure. In this context, it is important to note that the data processing for SaaS can be executed on different physical resources such as a server as well as a hardware accelerator with different performance and power consumption. To this end, a very significant feature of heterogeneous CDCs is that they offer the flexibility of meeting user demands for SaaS by choosing among the available physical resource alternatives. To utilize this flexibility, a CDC resource manager must decide which resource alternative will be chosen, along with the decision of the physical resource the request will be assigned to. In this thesis we propose ACCLOUD-MAN (ACCelerated CLOUD MANager), a novel resource manager for heterogeneous CDCs. ACCLOUD-MAN’s resource management objective is to reduce the power consumption of the CDC in order to support green computing. To this end, the resource allocation problem is modeled as an integer linear programming problem and is implemented in MATLAB, along with a cloud data center simulation platform. We evaluate the performance of ACCLOUD-MAN under different realistic cloud workloads. Simulation results show that the proposed ACCLOUD-MAN outperforms existing resource allocation methods such as OpenStack.