Research Of Routing And Flow Optimization On IP-Based Network

IP-based data center network is a special-line netw ork which is independent of the telecom operator’s existing underlying network. The coexisting of special-line network and telecom operator’s fundamental networks bring in new network architecture in which Shortest-Path mode and Best-Effort mode may bring in the problem of package losing, delay, jitter and so on. These are not qualified for high quality QoS traffic.This paper aims at solving the routing and traffic optimization problems in data center networks. Based on the research of existing multi-path traffic scheduling algorithm and multi-metric routing algorithm, MMMSA (maximize mixing metric scheduling algorithm) and AHP-WSA (Analytic Hierarchy Process Weighted Scheduling Algorithm) are proposed in this paper.This paper focus on two types of traffic:Aggregate Traffic and Explicit Traffic. The object of MMMSA is aggregate traffic. Different with ECMP (Equal-Cost Multipath Routing) which just balances the number of flows rather than the bit rates that is more likely to lead to link congestion, MMMSA considers the size of flow effectively and divides the aggregate traffic into two types:large flows and small flows. MMMSA achieves traffic optimization among multiple paths based on hybrid multiple metrics. The object of AHP-WSA is explicit traffic. AHP-WSA uses AHP (a method which is widely used in multiple attribute decision making) to establish multiple paths between source and destination with different weight. Then the business flows with different types will be scheduled to the corresponding paths.The proposed methods are verified in OPNET network simulation software. The results show that the traffic scheduled by these two algorithms achieves low end-to-end delay and high throughput. So the methods proposed in this paper cut down the overhead and improve the network performance effectively. It is significant for the further research of routing and traffic optimization problem in IP data center network.