POF缓存加速方案的优化方法

Optimization Method for POF Caching Acceleration Scheme

在软件定义网络领域,协议无关转发(POF)可编程数据平面能够快速地对新型网络协议进行支持,具有较高的实用价值。目前POF数据平面主要基于软件平台实现,性能较差。为了提高POF数据平面的性能,提出了基于缓存加速的方法。目前的缓存加速方案主要集中研究于快速转发路径的设计与性能优化,忽略了缓存失配路径上的额外开销。因此POF数据平面通过这些缓存加速方案要获得显著的转发加速效果,需要较高的缓存命中率。针对上述问题,本文基于最新的POF缓存加速方案提出了针对缓存失配路径的优化方法,通过改进缓存模块中的数据结构并将其传输到缓存失配路径,消除额外的字段提取开销,从而优化缓存方案整体的转发加速性能。本文在实际的POF可编程数据平面上实现了该方法并进行了对比分析。实验结果表明,本文方法在不影响原快速转发路径性能的同时,较好地提升了数据平面在现网流量中的转发性能。与原缓存方案相比,本文方法使数据平面获得显著转发加速效果,所需的缓存命中率降低了28.64%,同等缓存大小下吞吐最大提升了39.08%。

In the field of software-defined networks,the POF programmable data plane holds high practical value as it can quickly support new network protocols.However,the current implementation of POF data planes primarily relies on software platforms with poor performance.To enhance the POF data plane’s performance,various cache acceleration schemes have been proposed.Nevertheless,these cache acceleration schemes mainly concentrate on the design and performance optimization of the fast forwarding path,while overlooking the additional overhead on the cache-miss path.Consequently,a high cache hit rate is required to achieve significant cache acceleration on the data plane using these cache acceleration schemes.To address the aforementioned issues,this paper proposes an optimization method for the cache-miss path based on the latest POF cache acceleration scheme.By improving the data structure in the cache module and transmitting it to the slow forwarding data path,the extra field extraction overhead on the cache-miss path can be eliminated,thereby optimizing the overall acceleration performance of the caching scheme.The proposed method is implemented on an actual POF programmable data plane,and a comparative analysis is conducted.The experimental results indicate that our method enhances the forwarding performance of the data plane under real network traffic without affecting the performance of the original fast forwarding data path.Compared to the original caching scheme,our method decreases the cache hit rate required by the data plane to achieve significant forwarding acceleration 28.64%and improves throughput by a maximum of 39.08%for the same cache size.