دانلود رایگان مقاله مسیر یابی کارآمد برای اجرای سیاست middlebox در شبکه تعریف شده نرم افزاری

Abstract

Network applications require traffic to sequence through multiple types of middleboxes to enhance network functions, e.g., providing security and guaranteeing performance. Sequenced-middlebox policy routing on top of regular layer 2/3 flow routing is challenging to be flexibly managed by network administrators. In addition, various types of middlebox resources concurrently obtained by numerous applications complicate network-resource management. Furthermore, middlebox failures can lead to a lack of security and the malfunction of entire network. In this paper, we formulate a mixed-integer linear programming problem to achieve a network load-balancing objective in the context of sequenced-middlebox policy routing. Our global routing approach manages network resources efficiently by simplifying candidate-path selections, balancing the entire network and using the simulated annealing algorithm. Moreover, in case of middlebox failures, we design a fast recovery mechanism by exploiting the remaining link and middlebox resources locally. To the best of our knowledge, this is the first work to handle failures in sequenced-middlebox scenarios using OpenFlow. Finally, we implement proposed routing approaches on Mininet testbed and evaluate experiments’ scalability, assessing the effectiveness of the approaches. Results of the optimization on a test topology include an increase up to 26.4% of the throughput with respect to a sequenced shortest-path routing.

7. Conclusion

The middlebox policy makes the network routing problem more difficult. To solve this problem, we formulated an MILP optimization problem to allocate limited network resources and then solved the problem with the simulated annealing algorithm. The solution of the optimization problem indicates one path out of the candidate paths is assigned to each flow for balancing network loads. The global load-balancing routing not only balances network loads well but also keeps the number of flow entries on each switch within the range of its processing capability. Furthermore, we proposed fast local rerouting to tackle middlebox failures. The rerouting has no effect on the working subnetwork and can respond to middlebox failures quickly. With all these features implemented, we evaluated the efficiency and effectiveness of our approaches. Finally, our experiments on Mininet further validated our approach and attested the feasibility of applying our approaches to real networks. On our test topology, our LB approach shows an increase up to 26.4% on the throughput, when compared with the SP approach; and the LB approach also indicates expected end-to-end latency. Our fast local rerouting approach achieves similar results with the global rerouting approach: both approaches increase the throughput up to 16.8%, compared with the affected-flows-dropped approach. Our future work will include experiments on a real network testbed in addition to Mininet, and the design and evaluation of flow aggregation approaches.