Abstract
Cloud computing is growing fast and becoming more and more popular. The computing resources such as CPU and memory are becoming cheaper and the servers grow more and more powerful by the time. This enables clouds to host more virtual machines (VMs) than ever. As a result many modern data centers experience very high internal traffic inside the data centers due to the servers belonging to the same tenants communicating with each other. Since the modern VM deployment tools are not traffic-aware, the VMs with high mutual traffic often end up running far apart in the data center network and have to communicate over unnecessarily long distance. The resulting traffic bottlenecks negatively affect application performance and the network in whole and are posing important challenges for cloud and data center administrators. This thesis investigates how this problem can be resolved by consolidating VMs in clusters in different data center network architectures and deploy the produced clusters on the available server racks in a traffic-aware way. In order to achieve this the paper breaks the problem down in two parts. The VMs are consolidated with a VM clustering algorithm, successfully reducing the total cost of communication with 34 to 85\%, and the resulting clusters are assigned to the server racks with a cluster placement algorithm, which further reduces the total cost of communication with 89 to 99\%. The analysis shows that the optimization is done in a fast and computationally efficient way.
Cloud computing is growing fast and becoming more and more popular. The computing resources such as CPU and memory are becoming cheaper and the servers grow more and more powerful by the time. This enables clouds to host more virtual machines (VMs) than ever. As a result many modern data centers experience very high internal traffic inside the data centers due to the servers belonging to the same tenants communicating with each other. Since the modern VM deployment tools are not traffic-aware, the VMs with high mutual traffic often end up running far apart in the data center network and have to communicate over unnecessarily long distance. The resulting traffic bottlenecks negatively affect application performance and the network in whole and are posing important challenges for cloud and data center administrators. This thesis investigates how this problem can be resolved by consolidating VMs in clusters in different data center network architectures and deploy the produced clusters on the available server racks in a traffic-aware way. In order to achieve this the paper breaks the problem down in two parts. The VMs are consolidated with a VM clustering algorithm, successfully reducing the total cost of communication with 34 to 85\%, and the resulting clusters are assigned to the server racks with a cluster placement algorithm, which further reduces the total cost of communication with 89 to 99\%. The analysis shows that the optimization is done in a fast and computationally efficient way.