Testbeds and Research Infrastructure: Development of Networks and Communities. 9th International ICST Conference, TridentCom 2014, Guangzhou, China, May 5-7, 2014, Revised Selected Papers

Research Article

An Adaptive Fair Sampling Algorithm Based on the Reconfigurable Counter Arrays

Download98 downloads
  • @INPROCEEDINGS{10.1007/978-3-319-13326-3_35,
        author={Jing Wang and BingQiang Wang and Xiaohui Zhang and YunZhi Zhu},
        title={An Adaptive Fair Sampling Algorithm Based on the Reconfigurable Counter Arrays},
        proceedings={Testbeds and Research Infrastructure: Development of Networks and Communities. 9th International ICST Conference, TridentCom 2014, Guangzhou, China, May 5-7, 2014, Revised Selected Papers},
        proceedings_a={TRIDENTCOM},
        year={2014},
        month={11},
        keywords={Network traffic measurement Packet sampling Estimation error Reconfigurable parameter},
        doi={10.1007/978-3-319-13326-3_35}
    }
    
  • Jing Wang
    BingQiang Wang
    Xiaohui Zhang
    YunZhi Zhu
    Year: 2014
    An Adaptive Fair Sampling Algorithm Based on the Reconfigurable Counter Arrays
    TRIDENTCOM
    Springer
    DOI: 10.1007/978-3-319-13326-3_35
Jing Wang1,*, BingQiang Wang1,*, Xiaohui Zhang1,*, YunZhi Zhu1,*
  • 1: National Digital Switch System Engineering & Technology Center
*Contact email: Wangjingniu_2003@sina.com, wangbingqiang@ndsc.com.cn, zhangxiaohui@ndsc.com.cn, zhuwangzilz@163.com

Abstract

At present, how to trade off the balance between the memory resources and sampling accuracy balance has become one of the most important problems focused on by the network packet sampling algorithms. This paper discusses a novel adaptive fair packet sampling algorithm (AFPS) to solve the above problem by improving the use ratio of memory resources. The key innovation of AFPS is the reconfigurable counter structure composed of two counter arrays, by which the AFPS count the small flow and large flow in a differential way and the size of two arrays can be adjusted adaptively according to the dynamic flow size distribution. The reconfigurable counter structure ensures not only a high memory use ratio value under different network conditions but also accurate estimation of small flows so that the overall sampling accuracy of AFPS is improved. The theoretical analysis and evaluation on real traffic traces show that AFPS can estimate the small flows accurately and the estimation error of the large ones’ equals to SGS. Besides AFPS keeps the memory resource use ratio on almost 0.952 under different conditions so that it can use the memory resource efficiently.