Research On IP Network Traffic Engineering Based Upon Multi-Path Routing

Traditional web services, prevailing streaming media services and growing cloud computing services enforce a huge increase in traffic volume. The traffic increase with bursts frequently influences network performance, causing IP network providers to overprovision in their networks. However, this approach still cannot satisfy the demand from sharply growing customer traffic, while the network resources are not used efficiently. Therefore, traffic engineering is put forward for network optimization, and becomes a research focus.Routing optimization plays a key role in traffic engineering, finding multiple paths between source-destination pairs, and directing traffic on them. Optimized multi-path routing divides traffic flows flexibly between available paths, making network more efficient and robust. Today’s mainstream traffic engineering mechanisms in industry are reviewed; optimal load balancing model for multi-path routing are proposed; without knowledge of traffic matrix, multiple paths seeking and traffic distributing algorithms are proposed in ISP backbone network and data center network separately. The main contributions are as follows.(1) On the Development of Network Traffic Engineering. Methods employed in today’s traffic engineering are put together, and their implementations in IP network are classified for investigating from various aspects. First, public network data for research purposes, topology inclusive, are examined. Second, from an aspect of routing extent, intra-domain and inter-domain traffic engineering mechanisms are introduced; routing algorithms under these topics are also involved. Finally, new aspects are applied to review traffic engineering methods mentioned. Collecting and analyzing existing ones in various ways potentially lead to creative methods. Finally, traffic engineering in data center network is also reviewed.(2) Minimizing Sum of Path-cost Model and Algorithm for Traffic Balancing. Traffic balancing in routing optimization problem targets to minimize network congestion in traffic engineering. How to select paths for traffic balancing becomes a challenging problem. For minimizing network congestion, the paper argues that network congestion is determined by paths’congestion, and proposes the minimizing sum of path-cost model of traffic balancing. For path selection in routing traffic, a minimal cost path algorithm is proposed. On the basis of real network topology and traffic demand, experiments are conducted to verify the model and algorithm proposed, and results show a nearly20%decrease of maximal link utilization.(3) Min-Cut Multi-Path Routing Algorithm. Balanced traffic distribution can be archived through traffic engineering, and therefore network congestion is avoided. Today, routing through multiple paths to balance traffic distribution becomes a traffic engineering focus. As enforced by the restrictions of current routing mechanisms, existing multi-path routing algorithms could not get optimal traffic distribution, and therefore could not control network congestion efficiently. On the basis of minimum cut theory, the paper propose a min-cut multi-path (MCMP) routing algorithm, which select key paths of small quantity and distribute traffic evenly among selected paths. MCMP is apt to implement and control congestion at bottleneck links. With real traffic datasets, experiments are carried on European and North American backbone networks. Comparing to OSPF routing algorithm used in intra-domain network and a multiple paths routing algorithm used in optimal model, the maximum link load resulted from MCMP routing algorithm decreases over41%and20%separately.(4) Big-Flow Collision Avoidance Multi-Path Routing Algorithm in Data Center Network. To accommodate demand’s variation and network’s scalability, traffic engineering in data center has become an interesting focus in research field. As services supported by data center proliferate in diversity and in extent, big flows herein are popular; however, traditional routing algorithms cannot afford to resolve the collision problem of the flows. The paper propose an collision avoidance algorithm for the problem mentioned; the algorithm finds a path set in advance, primary routes for delivering big flows, and secondary routes for sharing traffic load simultaneously. Experiments carried out in typical data center topology exhibit that the proposed algorithm performs well in reducing maximum link load of core links, as opposed to traditional routing algorithms. Our collision avoidance algorithm is suited for deploying in distributed manner, since its nature of not concerning network status.