Reseach Of Several Significant Topics In Switch Network Based On Load-balancing Theory

The explosively rapid growth of information network,which is represented by TheInternet,making it a more important position of modern social life. In the consumptionof network resources and network congestion increasingly significant, how to make amore reasonable allocation of network resources, how to reduce congestion, and toguarantee the delay and improve the efficiency of existing equipment is challengingnetwork research. For the switching nodes of transport network, you need a higherexchange rate and network utilization.With the continual emergence of new needs,the patterns of traffic which carried bythe network has become increasingly diverse.And network traffic is also experiencingrapid growth,making the congestion of network getting worse.Load balancing,whichrecently has drawn a great deal of interest,is precisely raised to solve these problems.The idea of load balancing is widely used in various nodes in the network.Circuitswitch based on load balancing is to overcome the disadvantage that centralizingschedule resulting in non-uniform load distribution,and high complexity;In high-speedlarge-capacity multi-level packet switch network,load balancing mechanism is able toachieve100%throughput based on simplified scheduling policy,and have delayperformance guaranteed.In this paper,We first test the switch fabric performance with multiple traffic model.A traffic model is a simulation of the real traffic in network, characterized inrandomness, sudden current and self-similar.Traffic model implemented in this article isdivided into two parts:arrival models, and destiny distribution models. The arrivalmodels include traditional models, such as the Poisson, Bernoulli, exponentiallydistributed ON/OFF sources, and the self-similar model, heavy-tail ON/OFF source.And the destiny distribution models include uniform,diagonal and hotspot.Then we studied the semi-rearrangement characteristics in non-blocking circuitswitched networks,and the influence of flow aggregation on the load balancingperformance. We proposed a semi-rearrangeable network load balancing algorithmnamed SRN-LB,which perform load balancing when a new connection is established,and perform rearrangement when disconnecting a connection to maintain a balanced state, effectively improve the efficiency of network utilization.Finally, we extended the results to the field of packet switching,studied theinfluence of different flow aggregation strategy due to network utilization. Multi-stagenetworks based on crossbar is suitable for output-module flow aggregation,and moduleto module flow aggregation applies to shared buffer networks. The analysis shows thatthe larger the aggregated flow size, the better the traffic is balanced.