Flow Characteristic-Aware Cache Replacement Policy for Packet Processing Cache

Flow Characteristic-Aware Cache Replacement Policy for Packet Processing Cache
复制标题

DOI:
10.1007/978-3-030-03402-3_18
复制
发表时间:
2018-04
期刊:
Advances in Intelligent Systems and Computing
影响因子:
--
通讯作者:
Hayato Yamaki
Hayato Yamaki
中科院分区:
其他
文献类型:
--
作者:
Hayato Yamaki

文献摘要

相似文献

从数据包处理吞吐量和功耗的角度来看,互联网业务量的增加成为路由器面临的一个严重问题。分组处理缓存(PPC)是一种很有前途的方法,以满足需求。PPC可以通过将TCAM查找结果存储到高速缓存中并重用它们来减少对三进制内容可寻址存储器(TCAM)的访问次数,TCAM占路由器功耗的很大比例。对于PPC,该高速缓存未命中率对吞吐量和功耗具有显著影响。因此,减少高速缓存未命中的数量是PPC的主要关注点。在这项研究中,我们首先分析了网络中的大象流和老鼠流,揭示了PPC中的数据包行为,并提出了一种新的缓存替换策略的分析的基础上。本文提出的命中支配缓存(HDC)对大象流给予高优先级,对老鼠流进行快速驱逐。我们的模拟表明,与传统的4路LRU PPC相比,HDC可以将PPC中的缓存未命中数量减少29.1%。此外,我们估计HDC的硬件成本,通过使用Verilog-HDL和显示,它是可比的4路LRU,虽然HDC执行,如果该高速缓存是由8路组关联缓存。
The increase in internet traffic amount becomes a serious problem for routers from the aspects of the packet processing throughput and the power consumption. Packet processing cache (PPC) is a promising approach to meet the requirements. PPC can reduce the number of accesses to ternary content addressable memory (TCAM), which accounts for a large percentage of the power consumption of a router, by storing the TCAM lookup results into a cache memory and reusing them. For PPC, the cache miss rate has significant impact on the throughput and the power consumption. Thus, reducing the number of cache misses is a main concern for PPC. In this study, we first analyze the elephant flows and mice flows in networks to reveal the packet behavior in PPC and propose a novel cache replacement policy based on the analysis. Hit dominance cache (HDC), proposed in this paper, gives high priority to the elephant flows and evicts the mice flows rapidly. Our simulation showed HDC can reduce the number of cache misses in PPC by up to 29.1% compared to conventional 4-way LRU PPC. In addition, we estimated the hardware cost of HDC by using Verilog-HDL and showed that it is comparable to those of 4-way LRU though HDC performs as if the cache was composed of 8-way set associative cache.