چكيده به لاتين
Today using of traffic demand and Bandwidth requrment increased. Data Center Network (DCN) is one of the important infrastructure for response to this issue. In DC, exist much challenge, such as load balancing, energy consumption, performance and Quality of Services (QoS). The emergence of SDN, and consequently the separation of data Plain and control plain from each other, has significantly differentiated these challenges from the past. This thesis addresses some of these challenges, such as reducing power consumption and reducing TCAM memory usage.
Accordingly, an online/offline algorithm using evolutionary methods for equilibrium. The service charge is informed by the grid energy efficiency approach, which reduces the number of switches active in the grid. The process was done in offline phase inside the controller. The desired graph is modified and delivered to the online phase for routing a new flow. This modified graph is created using a minimal graph that guarantees connectivity across all client servers as well as the ant colony optimization algorithm.
The proposed algorithm is implemented in two phase online and offline using network emulator called Mininet and FloodLight Controller. The graph examined in this thesis, according to the classical Data centers are the 4th order of Fat-tree architecture.Finally, the proposed method is compared with the default controller mode that the dijkestra routing algorithm uses. The results obtained on different traffic and interconnection scenarios show that the algorithm performs well in high congestion with streams with different origin and average volume.
For example, the component of the standard deviation of links defined to evaluate load balances has improved by more than 30% compared to the controller. In addition, energy Consumption in the proposed algorithm is reduced by 25%. These are the delays and bandwidth due to satisfying service quality for components.
