Şebeke fonksiyonları sanallaştırma kullanarak IMS şebekelerinde dinamik yük paylaşımı

Long Term Evolution (LTE) access technology has become a de facto standard to meet the unprecedented mobile traffic increase. LTE, which has an all IP based architecture, uses IP Multimedia Subsystem (IMS) at the core network for delivering and managing IP multimedia services. IMS systems has several functions like P-CSCF, I-CSCF, S-CSCF and HSS. On the other hand cloud technologies and commercial off the shelf(COTS) HW are evolved in recent years which gives also the possibility to virtualize network functions. Virtualization at telecom networks is also necessary to overcome high capex investment costs coming with high data traffic due to smartphones, Internet of Things(IOT) devices and decreasing revenues. Network Function Virtualization (NFV) is utilized to simplfy the deployment and management of constantly evolving and highly complex networking services. NFV also helps to dynamically scale network resources when capacity is really needed. Using COTS HW will also reduce opex costs since the HW used for IT and networks services will be similar. On the other hand NFV gives the possibility to decrease used power resources by allowing dynamic scaling down the capacity when the network traffic is low. This also makes the green network in reality. In this thesis, we propose a dynamic load management framework for IMS networks using NFV. In the proposed framework, IMS functions are created within a single virtual machine (VM) instance and moved to the cloud environment hosted by a general purpose hardware system. Requests coming from the IMS clients are first processed by a load balancer module to efficiently distribute the incoming load over a pool of multiple IMS VMs. The decision of switching an IMS VM instance on or off is performed by the VM provisioning service using the resource utilization information periodically received from the IMS VMs. We utilized open source tools to implement the dynamic load management framework. The proof-of-concept experiments using a real testbed environment demonstrate that the proposed framework significantly increases the scalability of the IMS networks for the highly loaded traffic scenarios. In thesis we have build a real testbed environment with open source tools. Open IMS is used for IMS functions like P-CSCF, I-CSCF, S-SCSF and HSS. Open SIPS is used as load balancer for balancing Sesssion Initiation Protocol (SIP) traffic towards IMS. IMS Bench SIPp used for SIP traffic generation. Zabbix monitoring tool used for monitoring IMS instances and starting new instances when needed because of high CPU load in active IMS nodes. The advantage of our testbed is showing real results since it is build completely by real products other than simulation tools like NS2 or Omnet+. These open source products installed on VMWare Workstation as VMs and necessary configurations done for these tools. Some configuration and source code change was also necessary at Open IMS, Open SIPS side for the integration of these tools. Scripting at Open SIPS load balancer and IMS Bench SIPp was necessary for the test scenarios like IMS calls , IMS registration and distributing the SIP traffic to IMS nodes effectively. To obtain the test results we run scenarios with LB and without LB for several times and used average values in our graphics. Results obtained shows that putting load balancer is increasing CSRs (Call Success Rate) significantly when compared with the scenario without load balancer.