The Rearch On Network Energy Consumption Model Based On Multi-commodity Flow And Intelligent Algorithm

In recent decades, greenhouse effect which is resulted from carbon emission has be-come increasingly prominent. Various sectors have developed a consensus of the im-portance of energy conservation and low-carbon economy. The past few years have wit-nessed the rapid development of information technology. However, the amount of carbon emission from IT industry has increased markedly. The research data indicates that the ICT sector accounts for approximately 2% of carbon emission produced by all manufac-turing industries, although the figure is similar to that in global aerospace industry, it in-creased faster than any other sectors. Moreover, in developed countries such as the United Kingdom, the figure even reached 10%, and the growing trend will continue in the future.Most of time, the bandwidth utilization is less than 40% in real network due to the burst and periodicity. However, it is a fact that the energy consumption of network equip-ment is independent of workload. The power consumed under lower utilization is even equal to that in peak-hour. The conception of Green Network is proposed under the cir-cumstances. In the view of engineering, the main ideal of Green Network is to minimize the energy consumption while satisfying the requirement of bandwidth and QoS. There have been various researches on this area, and according to the scope the optimization, the approaches can be divided in to two levels—device level and network level.. Energy consumption optimization of device level is towards the individual devices, such as rout-ers, switches, line cards, etc. and the aim is to make energy consumption on devices in proportion to workload. The proposed methods are DVS, link adaptation strategy, exten-sible components, traffic forecasting, etc. The aim of network level optimization is to make the consumption of power in proportion to network workload and it has been real-ized through Energy-Aware Routing (EAR). The problem has been classified into capac-itated multi-commodity flow problem (CMCF), which is NP-complete.The basic idea of CMCF is to aggregate traffic into a network subset and shut down (or sleep) other network elements, which makes the total energy consumption in propor-tion to workload on the whole internet. The aim of CMCF is to find the small-sized net-work topology with lowest energy consumption. CMCF problem can be solved by clas-sical mathematic model and in this paper we aggregate same requirements in order to decrease the number of variables and speed up the solution process. However, since the mixed integer programming (MIP) is NP-hard, computing time in the topology of large scale is unacceptable. As a result, we proposed an Clone Ant-based Ant Colony Optimi-zation Routing Algorithm (CACO-RA). In our algorithm we generate a kind of phero-mone for a distinct destination and try to maximize the traffic aggregation to fewer nodes and links. Besides, what we implement is a split flow scheduling solution, which takes full advantage of the bandwidth of the network. And simulation test results with random topologies show that our algorithm have better performance not only in power consump-tion、computation time but also in reject rate.、Although the performance of split flow scheduling solution is quite fair, it brings another problem, that is delay. The delay of traditional algorithm is undoubtedly the smallest, since it is based on the shortest path, and due to the single path transmission of flow traffic, there is no jitter problem. In CACO-RA, the algorithm we allocate multiple paths for each demand, which brings the problem of delay and jitter. In order to get a trade-off between energy consumption and QoS, we modified CACO-RA combined with the idea of particle swarm optimization and name the new algorithm as Hybrid Ant Col-ony Optimization (HACO). In HACO, the input of each particle is the output of CACO-RA, and the fitness of each particle is determined by the QoS and the load balance. The particles learn from the combined subset after every iteration. After multiple iterations, we can finally get a network subset with lower energy consumption, less delay and better load.