OpenTM这篇文章介绍了如何使用Openflow提供的控制转发分离结构以及本身自带的统计功能,来测量网络的流量矩阵。流量举证的测量一直以来存在很多困难。这篇文章结合Openflow给出了一种准确、快速的测量方式。在统计流量信息时,只需要10次左右的询问就能统计出所需的流量矩阵。
以下是论文里的一些精彩的片段。

1. Estimating the point-to-point TM in a network is an essential part of many netowrk design and operation tasks such as capacity planning,routing protocol configuration,network provisioning,load balancing and anomaly dectection.关于流量感知的5个方面.
2. 现有IP网络里流量感知的困难:Direct and precise measurement of TM in large IP networks is extremely difficult, if not infeasible, due to the large number of OD pairs, the high volume of traffic at each link, and the lack of measurement infrastructure.[1-Robust traffic matrix estimatin with imperfect information]总结起来就是三点,large number OD pairs,high traffic at each link,lack of measurement infrastructure.
3. 测量TM来自三种方式:indirectly from link loads,directly from link loads,both of them.直接测量可以使用思科的netflow。由于缺乏测量机制,直接测量会带来巨大的损耗。
4. 关于Openflow有用一段文字:In this paper, we revisit the TM estimation problem using direct measurements in the context of OpenFlow-based networks. OpenFlow is an open standard that makes any changes to the network control plane very easy by separating the data and contol plane. An OpenFlow network consists of OpenFlow switches(data plane) managed by a logically centralized OpenFlow controller move the prohibituve cost of direct measurements for TM estimation. Unlike commodity swithches, flow level opeartions are streamlined into OpenFlow switches which lets us query for flow statistics, enabling access to accurate flow statistics.
5. 可借用的一些句势:Our work shows the possibility of directly measurement of TM with the least overhead as long as the infrastructure(OpenFLow here) provides the appropriate feature set for measurements. We note that the scope of our work is limited to the networks where OpenFLow can be deployed,i.e.,where maintaining per-flow counters is likely tractbale.
6.  值得借鉴的思想:在设置询问对象时,如何做决策?应当采取什么样的策略?OpenTM这篇文章中进行了一些初步的讨论和和研究。Even though all these schemes result in TM estimations with very small errors, our analysis and experiments and the maximum query load on individual switches, i.e., the perfect query distribution sometimes results in the worst estimate, and the best estimation can lead to the worst query distrubition amongst switches.
7. 可以借鉴的句子:Even though using a small testbed for evalution has its own shortcomings, we believe that most results would not be significantly different in larger networks. 合理的为自己开脱。
8. The contributions of this work are two-fold. First,….Second,….
9. 表述自己的想法:We explore the idea of ..To the best of our knowledge,…
10.  举得有数据的例子:It is conmmonly believed that the number of concurrently active flows in large enterprise IP networks is small. According to the data from the 8000-host network at LBNL, the total number of active flows in their network never exceeds 1200 in any second. The data from the Stanford Computer Science and Electrical Engineering network with 5500 active hosts shows that their number of active flows stays well below 10000. Currently, our system generates a single query for a single source-destination IP pairs. As an improvement, a single query can be generated for all flows sharing the same path, as long as the IP address could be aggregated.
11. 规避一些不该有的工作:The relation between an efficient querying frequency and flow and network dynamics is outside the scope of this work.
12. 两个有用的工具:We use the NetwEm to emulate network dyaly and loss, and use Iperf to generate the input traffic.

2014.12.8—关于wordpress发表文章没有空格问题的解决办法

在主题的function.php里面添加: add_filter('tiny_mce_before_init', 'preserve_nbsp_chars'); function preserve_nbsp_char...

阅读全文

2014.12.8—关于微博的内容分发机制的小调研

         今天在思考关于ICN的pull方式,以及现有的CS等通信机制的问题。关于Pull和Push两种机制,我在今天的另外一篇博客中已经有...

阅读全文

2014.12.8—Pull和Push通信机制的研究

所谓PUSH技术是一种基于客户服务器机制,由服务器主动的将信息发往客户端的技术。 同传统的拉技术(PULL)相比,两者最为主要的区别在于前者的是由服务器主动发...

阅读全文

4 条评论

欢迎留言

此站点使用Akismet来减少垃圾评论。了解我们如何处理您的评论数据