| 摘要: |
| 基于通用多核的网络转发性能难以满足高速网络流量线速处理的需求.软硬件结合的异构网络处理平台以其较高的性能和灵活性在网络处理领域得到广泛应用，但是如何基于异构平台实现高效的路由查表算法仍需进行深入研究，多核资源利用率低、共享冲突严重和访存次数多的问题是制约传统路由查表算法在异构网络处理平台实现性能提升的主要问题.为此，基于异构网络处理平台（network processing platform，简称NPP)提出一种可配置并行路由查表机制（configurable parallel lookup，简称CPL).CPL中的多线程并行查找和路由表的多副本存储技术在提高多核资源利用率的同时，实现了零冲突访问路由表项.此外，考虑到不同场景下路由前缀分布的差异，CPL支持通过配置对多级路由表的组织结构进行调整，从而有效地减少了路由表访问次数.最后在NPP上，对CPL和传统的查表算法进行性能测试和对比，验证了CPL的可用性和高效性. |
| 关键词: 多核 并行 零冲突 可配置 |
| DOI： |
| 分类号: |
| 基金项目:国家高技术研究发展计划（863)（2015AA016103)；高性能计算协同创新中心优秀研究生创新资助项目 |
|
| Research on Reconfigurable Parallel Routing Lookup Algorithm Based on Heterogeneous Network Processing Platform |
|
YAN Jin-Li, LÜ Gao-Feng, TANG Lu, LI Tao, SUN Zhi-Gang
|
|
School of Computer, National University of Defense Technology, Changsha 410073, China
|
| Abstract: |
| The forwarding performance of general-purpose multi-core server cannot satisfy the demand of wire-speed processing of high-speed traffic very well. Although the heterogeneous network processing platforms combining software and hardware have been applied widely in network processing field for their high performance and flexibility, the problem of designing high-efficiency routing table lookup algorithm in this platform still needs further study. The main problems restricting the performance improvement of heterogeneous network processing platform include low utilization of multi-core resources, serious shared-resources access conflict and large amounts of memory access. Thus, this paper proposes a routing lookup mechanism named CPL (Configurable Parallel Lookup) based on heterogeneous network processing platform NPP (Network processing Platform). The technologies of multi-thread parallel lookup and multi-copy management of routing tables in CPL improve the utilization of multi-core resources and accessing routing tables with zero-conflict. In addition, given that the distribution differences of routing prefix in different scenarios, CPL regulates the structure of multi-level routing table by configuration, which reduces the number of routing table accesses efficiently. At last, after making the performance comparison of CPL and traditional lookup algorithm, the experiments demonstrate the availability and high-efficiency. |
| Key words: multi-core parallel zero-conflict configurable |
