云基础设施中TPL2-SW网络模型的优化

为解决传统树型网络拓扑对大规模云集群造成的性能瓶颈,针对一种新的网络模型TPL 2-SW进行优化,提出了多路径负载均衡路由策略(MCLBR)和转发路由的硬件加速方案。通过在节点间建立网状链路,将负载指数纳入路由判定因素,并允许节点间负载分流;同时基于FPGA实现硬件转发路由器,并验证了在二层语义对多跳传输加速的可行性。仿真实验证明,优化方案可以在TPL 2-SW基础上显著提高网络的吞吐量和传输效率。

Optimization for TPL2-SW networks of cloud infrastructure

To break through the performance bottlenecks of the large-scale cloud infrastructure caused by the traditional tree-network topology, this paper proposed the MCLBR, a multi-channel load balancing routing strategy based on a new networks model TPL2-SW. The MCLBR established mesh links between the nodes and took the node load status into the routing decisions in order to improve the throughput and the redundancy, besides data flows could be redirected to spread the traffic. To reduce the delay of the multi-hop, a hardware router, which worked at the ethernet lay-2 semantics, was implemented based on FPGA. Simulation and experiment result shows the TPL2-SW network can be improved significantly by the optimization.