在此將會實驗位元圖交集法、R*-Tree 分類法以及 R*-Tree 與位元圖 混合法在處理不同規則表與不同規則表大小時封包比對時記憶體存取 次數，以及記憶體的需求。主要的目的是比較我們的方法與其他的方 法在效能上的差異。規則表的種類為先前所提到的 acl、fw、和 ipc，
針對每種規則表都會測試不同數量的規則對效能的影響為何。詳細的 實驗參數如表 2 中所示
表 2 不同種類規則表實驗參數
參數 數值
規則數量 1k~10k
測試封包數量 100k
規則表種類 acl, fw, ipc
規則的維度
節點中 MBR 最大數量
5 64
節點中 MBR 最小數量 32
記憶體頻寬 32 bit
29
956 1846 2562 3685 4568 5569 6659 7727 8788 9841
Memory Access per Packet
Rule Number
Memory Access (acl)
Bitmap Intersection R*-Tree
R*-Tree + Bitmap Theoretical Bitmap Intersection Theoretical R*-Tree Theoretical R*-Tree + Bitmap
0
956 1846 2562 3685 4568 5569 6659 7727 8788 9841 Memory usage (KB based on log2)
Rule Number
Memory Usage (acl)
Bitmap Intersection R*-Tree R*-Tree + Bitmap
30
813 1739 2709 3760 4703 5661 6581 7516 8463 9389
Memory Access per Packet
Rule Number
Memory Access (fw)
Bitmap Intersection R*-Tree
R*-Tree + Bitmap Theoretical Bitmap Intersection Theoretical R*-Tree Theoretical R*-Tree + Bitmap
0
813 1739 2709 3760 4703 5661 6581 7516 8463 9389 Memory usage (KB based on log2)
Rule Number
Memory Usage (fw)
Bitmap Intersection R*-Tree R*-Tree + Bitmap
31
978 1921 2899 3786 4758 5711 6650 7600 8563 9511
Memory Access per Packet
Rule Number
Memory Access (ipc)
Bitmap Intersection R*-Tree
R*-Tree + Bitmap Theoretical Bitmap Intersection Theoretical R*-Tree Theoretical R*-Tree + Bitmap
0
978 1921 2899 3786 4758 5711 6650 7600 8563 9511 Memory usage (KB based on log2)
Rule Number
Memory Usage (ipc)
Bitmap Intersection R*-Tree R*-Tree + Bitmap
32
33
34
35
圖 18 Response time of different node size with rule size 1000
圖 19 Tree height of different node size with rule size 1000
0
Memory Access per Packet
Node size
Memory Access (size 1000)
ACL FW IPC
Height of tree
Node size
Height of tree (size 1000)
ACL FW IPC
36
圖 20 Memory Access of different node size with rule size 1000
圖 18、圖 19 與圖 20 分別是在規則數量為 1000 筆時，不同節點大
Number of traversed node
Node size
Traversed node (size 1000)
ACL FW IPC
37
圖 21 Memory Access of different node size with rule size 10000
0
Memory Access per Packet
Node size
Memory Access (size 10000)
ACL FW IPC
38
圖 22 Tree height of different node size with rule size 10000
圖 23 Traversed node of different node size with rule size 10000
0
Height of tree
Node size
Height of tree (Rule size 10000)
ACL FW IPC
Number of traversed node
Node size
Traversed node (Rule size 10000)
ACL FW IPC
39 只會使用 Source / Destination IP address 這兩個欄位來設定布隆過濾器，
40
41
956 1846 2562 3685 4568 5569 6659 7727 8788 9841
Memory Access per Packet
Rule Number
Memory Access (acl)
R*-Tree + Bitmap + Bloom R*-Tree + Bitmap Multiple R*-Tree + Bitmap
0
813 1739 2709 3760 4703 5661 6581 7516 8463 9389
Memory Access per Packet
Rule Number
Memory Access (fw)
R*-Tree + Bitmap + Bloom R*-Tree + Bitmap Multiple R*-Tree + Bitmap
42
978 1921 2899 3786 4758 5711 6650 7600 8563 9511
Memory Access per Packet
Rule Number
Memory Access (ipc)
R*-Tree + Bitmap + Bloom R*-Tree + Bitmap Multiple R*-Tree + Bitmap
43
綜合以上實驗結果，在封包處理速度上，我們所提出的方法在規則 表小的時候有著與位元圖交集法有著相當的效能。在處理規則數量較 大的規則表時，我們所提出的方法效能會比其他的演算法更好。此外，
我們所提出的方法所需要的記憶體空間是遠勝於位元圖交集法的，所 以在封包處理的過程中，我們所提出的方法需要存取記憶體的次數比 位元圖交集法要來的少。整體來說我們的演算法無論是在比對速度或 是記憶體的使用上都優於 R*-Tree 演算法以及位元圖交集法。另外我們 所提出用來改善我們方法的演算法中，從結果來看多重 R*-Tree 確實是 能夠讓比對的效能更加上升。
44 利用 ClassBench 這套能反映現實情況的工具來實驗我們的方法在不同 特性的規則表下的效能。實驗結果指出，我們的方法在比對效能上平
45
參考文獻
[1]
A. Cuttman, “R-Tree: A Dynamic Index Structure for Spatial Searching,” in proceedings of SIGMOD, pages 47-57, 1984.[2]
B. Seeger, H.-P. Kriegel, N. Beckmann, and R. Schneider, “The R*-tree: An Efficient and Robust Access Method for Points and Rectangles,” in proceedings of SIGMOD, pages 322-331, 1990.[3]
C. Maindorfer and T. Ottmann, “Is the Popular R*-Tree Suited for Packet Classification?,” in proceedings of IEEE NCA, pages 168-176, 2008.[4]
D. Taylor, “Survey & Taxonomy of Packet Classification Techniques,”ACM Comput. Surv., vol. 37, no. 3, pages 238–275, 2005.
[5]
D.E. Taylor and J.S. Turner, “ClassBench: A Packet Classification Benchmark,” in proceedings of IEEE INFOCOM, 2005.[6]
G. Varghese, “Network Algorithmics: An Interdisciplinary Approach to Designing Fast Networked Devices,” Morgan Kaufmann publishers, 2005.[7]
H. J. Chao and B. Liu, “High Performance Switches and Routers,”Wiley-IEEE Press, 2007.
[8]
M.H. Overmars and A.F. van der Stappen, “Range Searching and Point Location Among Fat Objects,” Journal of Algorithms, vol. 21, no. 3, pages 629-656, November 1996.[9]
P. Gupta and N. McKeown, “Algorithms for Packet Classification,”IEEE Network, vol. 15, no. 2, pages 24-32, March/April 2001.
[10]
P. Gupta and N. McKeown, “Packet Classification using Hierarchical Intelligent Cuttings,” in proceedings of IEEE Micro, vol.20, no.1, pages 34-41, January /February 2000.[11]
S. Singh, F. Baboescu, G. Varghese, and J. Wang, “Packet46
Classification using Multidimensional Cutting,” in proceedings of SIGCOMM, pages 213-224, 2003.
[12]
T.V. Lakshman and D. Stiliadis, “High-Speed Policy-based Packet Forwarding Using Efficient Multi-dimensional Range Matching,” in proceedings of SIGCOMM, pages 191-202, 1998.[13]
C.R. Hsu, C Chen and C.Y. Lin, “Fast Packet Classification using Bit Compression,” in proceedings of IEEE Global Telecommunication Conference, 2005.[14]
P. Cupta and N. McKeown, “Packet Classification on Multiple Fields,”in proceedings of Sigcomm, Computer Communication Review, pages 147-160, September 1999.
[15]
F. Baboescu, S. Singh and G. Varghese, “Packet Classification for Core Routers: is there an alternative to CAMs,” in proceedings of Infocom, vol. 1 pages 53-63, March 2003.[16]
W.T. Chen, S.B. Shin and J.L. Chiang, “A Two-stage Packet Classification Algorithm,” in proceedings of Conference on AINA, page 762-767, March, 2003.[17]
X. Sun and Y.Q. Zhao, “Packet Classification using Independent Sets,”in proceedings of ISCC, vol. 1, page 83-90, June/July 2003.