透過您的圖書館登入
IP:18.118.144.69
  • 學位論文

多核心川流繪圖處理器應用於網路流量變異偵測之實現與研究

The Implementation of Sketch-based Network Traffic Change Detection on Multi-Core Stream Processor

指導教授 : 賴裕昆
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


網路流量變異偵測被用來探測網路異常行為的發生,然而隨著網路系統越趨複雜,單一觀測點的系統已無法提供完整的網路資訊。藉由增加觀測點的數量能使偵測系統取得較完整的網路概觀,但單位時間所需處理的資料量也隨之增加。為了能在相同的時間內處理倍數增加的資料量,本論文採用多核心繪圖處理晶片作為網路流量變異偵測系統之運算加速器。 Storm-1川流處理器提供單一指令同步處理多組資料(Single Instruction Multiple Data, SIMD)的處理架構與超長指令字組(Very Large Instruction Word, VLIW)的執行方式。這樣的處理器架構相當適合用來處理經由網路觀測器所統整出來的資料陣列,利用其平行運算的特性與多階層式的記憶體架構,能提供更充裕的運算資源,藉以分析大規模網路環境下的資料。 本論文提出川流式流量變異偵測系統,藉由改變Krishnamarthy等人所實現的K-ary Sketch流量變異偵測系統的執行順序,以降低非必要性資料傳輸所耗費的記憶體存取時間,以及取出變異數過程中對同一筆資料多次運算的問題。川流式流量變異偵測系統能以低功耗的Storm-1川流處理器快速完成複雜的雜湊函數值計算,並在數毫秒內完成對異常封包流的判斷。相較於原本K-arySketch流量變異偵測系統,本系統能有相當顯著的效能提升。

並列摘要


In this dissertation, we implement a sketch-based network traffic change detection system for IP networks in a low-power SIMD stream processor. The design is based on the K-ary sketch proposed by Krishnamarthy. Leverage on special three-tier memory architecture, the sketch data structure is organized internally in a form of data stream to hide the memory access latency with kernel processing time. We explore the design on stream architecture and conduct cycle-based simulation with real-world Internet traffic traces. The system is capable of processing sketch data sent from many observation points in a large scale of network. It can also identify the flow ID exceeding the predefined threshold very efficiently in real time.

參考文獻


[28] 魏特佑 (Theophilus Wellem). 於通用繪圖處理器架構實現川流式網路封包處理之研究與探討 (stream-based packet on general purpose gpu architecture). Praca magisterska, 中原大學電子工程研究所, 2011.
[1] Jung Ho Ahn, W. J Dally, B. Khailany, U. J Kapasi, A. Das. Evaluating the imagine stream architecture. strony 14-- 25, Czerwiec 2004.
[4] Xiaowen Chu, Kaiyong Zhao, Mea Wang. Accelerating network coding on many-core GPUs and multi-core CPUs. Journal of Communications, 4(11), Grudzie/n 2009.
[5] Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest, Clifford Stein. Introduction to Algorithms. The MIT Press, wydanie third edition, Lipiec 2009.
[7] Sangjin Han, Keon Jang, KyoungSoo Park, Sue Moon. PacketShader: a GPU-accelerated software router. Proceedings of the ACM SIGCOMM 2010 conference on SIGCOMM, strony 195--206, New Delhi, India, 2010. ACM.

延伸閱讀