Performance Implications of Problem Decomposition Approaches for SDN Pipelines
Performance Implications of Problem Decomposition Approaches for SDN Pipelines
复制标题
DOI：
10.1109/globecom42002.2020.9322392
复制
发表时间：
2020-12
期刊：
影响因子：
--
通讯作者：
William Brockelsby;R. Dutta
中科院分区：
文献类型：
--
作者：
William Brockelsby;R. Dutta
软件定义网络（SDN）允许组织以编程方式修改网络，以实现自定义转发行为并对不断变化的条件做出反应。虽然有许多方法可用于实现SDN，但那些利用转发表抽象（如OpenFlow和P4）的方法需要开发人员将问题分解为与可定义管道相关联的一个或多个表。本文通过分析对硬件和软件数据平面（包括通过使用SmartData利用硬件加速的软件数据平面）的性能影响，探讨了与不同问题分解选项相关的表深度和管道长度之间的权衡。
Software defined networking (SDN) allows organizations to modify networks programmatically to implement custom forwarding behavior and to react to changing conditions. While there are many approaches available to implement SDN those that leverage forwarding table abstractions such as OpenFlow and P4 require developers to decompose problems into one or more tables associated with a definable pipeline. This paper explores tradeoffs between table depth and pipeline length associated with different problem decomposition options by analyzing the performance impact on hardware and software data planes including software data planes leveraging hardware acceleration through the use of SmartNICs.