Using the load-balance mechanism in a multihoming network does not need to exchange lots of routing information to every connected ISP, and it is not necessary to conduct a lot of measuring.
In this study our contribution is to categorize the general load-balance algorithms and describe them by five generic parameters. We have reported these parameters and their effect in performance at different workloads. Therefore, when applying a load-balanced algorithm in a multihoming network, a suitable algorithm can be chos of the specific network. If the load-balanced th hp d nate y tr c at th iles, the CSN-type or ST-type algorithm ld ns d to crease measuring cost. If the traffic workload
ontrol of a narrower bandwidth should be included in the planning process f a multihoming network.
The performances of algorithms are compared at various workloads to observe eir responses to traffic congestion. Algorithms respond differently at light and heavy orkloads. When the workload is light, an algorithm that can detect the bottleneck to void the traffic-congested path yields a better throughput. Conversely, when the
orkload is heavy, the highest throughput is achieved by algorithms with the eighted dispatching scheme. However, utilizing a congested path to gain throughput nder a heavy workload leads to transmission delay. Finally, the algorithm with both
en to meet the characteristics
roug ut is omi d b affi e last m cou be co idere de the
is heavy, the weighed dispatching scheme is a good choice to gain higher utilization.
Using multiple economic links to gain an aggregated throughput is applicable in an enterprise network. However, according to our experiments, the consideration of the flow c
congestion detection and weighted dispatching yields both a better throughput and BU at various workloads.
In this study we also propose a per-c
ave been proven. Comparing to the measuring method using extra packets to
E
onnection timely detection scheme for end-to-end transmission, called WSDM. Its resource usage efficiency and ability to keep away from the outage path h
get end-to-end round trip time to do path selection for every connection, WSDM can achieve the same successful percentage of end-to-end transmission. WSDM can also provide better bandwidth utilization in a heavy workload situation owing to its weighted dispatching scheme.
The main benefit of WSDM to equipment vendors is its lower resource usage of measuring operations that can be implemented in cost effective hardware. For ISPs, WSDM consumes little bandwidth resource and does not require routing information exchange or the input traffic burden of measuring traffic. For an enterprise, WSDM provides an enhanced utilization of bandwidth.
52
Reference
1. A. Akella, B. Maggs, S. Seshan, A. Shaikh, and R. Sitaraman. A
ch 2004
measurement-based analysis of multihoming. In Proc. of ACM SIGCOMM, August 2003.
2. D. Goldenberg, L. Qiu, H. Xie, Y.R. Yang and Y. Zhang, Optimizing Cost and Performance for Multihoming, in Proceedings of the 2004 ACM SIGCOMM Conference, August 2004
3. F. Guo, J. Chen, W. Li, T. Chiueh, “Experiences in Building a Multihoming Load Balancing System,” In Proc. IEEE INFOCOM, Mar
4. K. Egevang and P. Francis, “The IP Network Address Translator(NAT),” RFC 1631, May 1994
5. http://www.radware.com 6. http://www.f5.com
7. http://www.deansoft.com.tw/Ehome.htm
8. A. Akella, S. Seshan, A. Shaikh, “Multihoming Performance Benefits: An Experimental Evaluation of Practical Enterprise Strategy, “ USENIX 2004 T. Bates, Y. Rekhter, “Scalable Support for Multi- homed Multi-provider Connectivity,” RFC2260, January 1998
Y. Rekhter, T. Li, “An Architecture for IP Address Allocation with CIDR,”
RFC1518, September 1993
teway Protocol 4 (BGP-4),” 1995, RFC1771, March 1995
C. Labovitz, R. Malan, and F. Ahanian, “Internet routing instability,”
IEEE/ACM Trans. Networking, vol.6, mo. 5, pp.515-558, 1998.
9.
10.
11. Y. Rekhter, T. Li, “A Border Ga
12.
53
13. C. Labovitz, R. Malan, and F. gins of Internet routing instability,”
14.
a network failure,” in Proc. International Symposium on
15.
01
n Proc. IEEE
17. main Routing: Problems and Solutions“, Technical Report, ork, Feb 2003.
ja, A. Bose, and F. Jahanian. “Delayed Internet routing 0, Stockholm, Sweden, pp. 175-187,
19. ilient
ciples
the ts on Reactive Routing,” ACM SIGMETRICS, San
21. D. Gan, “Load Sharing using IP Network Address Translation,”
Ahanian, “Ori in Proc. IEEE INFOCOM, 1999
C. Labovitz, A. Ahuja,and F. Ahanian, “Experimental study of Internet stability and wide-are
FaultTolerant Computing, June 1999
T.G. Griffin and B.J. Premore, “An Experimental Analysis of BGP Convergence Time,” in Proc. of ICNP 20
16. D. Pei, X. Zhao, L. Wang, D. Massey, A. Mankin, S. Felix Wu, and L. Zhang,
“Improving BGP convergence through consistency assertions“, i INFOCOM, 2002.
D.G. Andersen, H. Balakrishnan, M.F. Kaashoek, and R. Morris, “Res Overlay Networks,” in 18th ACM Symposium on Operating Systems Prin (SOSP), October 2001.
20. N. Feamster, D.G. Andersen, H. Balakrishnan, M.F. Kaashoek, “Measuring Effects of Internet Path Faul
Diego, CA, June 2003.
P. Srisuresh,
RFC 2391, August 1998
54
22. A. Feldmann, “Characteristics of TCP connection arrivals,” in: Self-similar Network Traffic and Performance Evaluation, eds. K. Park and W. Willinger,
23.
ng, June 1995
s. “ IEEE/ACM Transactions on Networking,
25.
ling phenomena, “ Computer Communications Review, vol.
John Wiley and Sons, pp. 367-399, 2000
V. Paxson, and S. Floyd, “Wide-Area Traffic: the Failure of Poisson Modeling,”
IEEE/ACM Transactions on Networki
24. M. Crovella and A. Bestavros. “Self-similarity in World Wide Web traffic:
Evidence and possible cause 5(6):835--846, Nov 1997.
A. Feldmann, A. Gilbert, W. Willinger, and T. Kurtz, “The changing nature of network traffic: Sca
28, no. 2, April 1998.
55
Appendix A
ple of experimental raw data:Connection transferred status at workload L Exam
z 3(3,1)
z
41 192.168.100.1 1041 1099857471.76 0.002725 1.465922 50250 267.8029OK 1.47
1
42 .22 0.000697 1.339441 50250 293.091OK 2.81
2 192.168.100.3 1002 1099857473.32 0.039352 1.629028 50250 240.9892OK 3.19
3
44 192.168.100.1 1044 1099857475.58 0.04981 1.479501 50250 265.345OK 5.31
4 475.95 0.008671 1.368072 50250 286.9572OK 5.56
45 192.168.100.1 1045 1099857477.06 0.000752 0.769431 50250 510.2188OK 6.07
46 192.168.100.1 1046 1099857478.06 0.000898 0.769562 50250 510.1319OK 7.07
47 192.168.100.1 1047 1099857479.07 0.000672 0.767956 50250 511.1987OK 8.07
48 192.168.100.1 1048 1099857480.07 0.000516 0.766338 50250 512.278OK 9.07
49 192.168.100.1 1049 1099857481.07 0.000731 0.735155 50250 534.0073OK 10.04
50 192.168.100.1 1050 1099857482.07 0.000709 0.753306 50250 521.1403OK 11.06
51 192.168.100.1 1051 1099857483.07 0.000704 0.771498 50250 508.8518OK 12.08
52 192.168.100.1 1052 1099857484.07 0.00067 0.769889 50250 509.9152OK 13.08
53 192.168.100.1 1053 1099857485.07 0.000725 0.768258 50250 510.9978OK 14.08
5 192.168.100.3 1005 1099857477.31 8.18344 50.136365 50250 7.830207OK 55.69
6 192.168.100.3 1006 1099857527.45 0.00073 0.769694 50250 510.0444OK 56.46
7 192.168.100.3 1007 1099857528.45 0.000619 0.768908 50250 510.5658OK 57.46
8 192.168.100.3 1008 1099857529.45 0.000665 0.767337 50250 511.6111OK 58.46
9 192.168.100.3 1009 1099857530.46 0.000645 0.765712 50250 512.6968OK 59.46
10 192.168.100.3 1010 1099857531.46 0.000602 0.76411 50250 513.7717OK 60.46
11 192.168.100.3 1011 1099857532.46 0.000683 0.772472 50250 508.2102OK 61.47
12 192.168.100.3 1012 1099857533.46 0.000641 0.770866 50250 509.269OK 62.47
13 192.168.100.3 1013 1099857534.46 0.000705 0.769265 50250 510.3289OK 63.47
14 192.168.100.3 1014 1099857535.47 0.000667 0.767662 50250 511.3945OK 64.47
54 192.168.100.1 1054 1099857486.08 8.38077 50.307779 50250 7.803527OK 64.62
55 192.168.100.1 1055 1099857536.38 0.000693 1.430508 50250 274.4327OK 66.06
15 192.168.100.3 1015 1099857536.47 0.006069 1.456644 50250 269.5086OK 66.17
z under traffic congestion situation using WMORBF algorithm
ID Src-IP Srcport Start time Setup Time Total time Size Rate State End time
192.168.100.3 1001 1099857471.76 0.021623 1.556455 50250 252.2258OK 1.56
192.168.100.1 1042 1099857473
43 192.168.100.1 1043 1099857474.56 0.059662 1.019458 50250 385.0851OK 3.83
192.168.100.3 1003 1099857474.94 0.000635 0.819395 50250 479.1073OK 4.01
192.168.100.3 1004 1099857
56
56 192.168.100.1 1056 1099857537.81 13 50250 510.0318OK 66.83
16 192.168.100.3 1016 1099857537.92 0.000575 0.819893 50250 478.8163OK 66.99
0.000635 0.769861 50250 509.9338OK 67.94
1 0 531.5237OK 68.91
19 192.168.100.3 1019 1099857540.93 0.000656 0.76661 50250 512.0963OK 69.94
2 .000678 0.764944 50250 513.2116OK 70.94
5 8.126867 49.616032 50250 7.912324OK 116.67
58 192.168.100.1 1058 1099857588.43 0.000743 0.769779 50250 509.9881OK 117.44
59 192.168.100.1 1059 1099857589.43 0.000659 0.769258 50250 510.3335OK 118.44
60 192.168.100.1 1060 1099857590.43 8.214796 49.6147 50250 7.912537OK 168.29
21 192.168.100.2 1021 1099857471.76 8.160236 181.6097 50250 2.161658OK 181.61
22 192.168.100.2 1022 1099857653.37 0.000755 0.764631 50250 513.4217OK 182.37
23 192.168.100.2 1023 1099857654.37 0.000651 0.754076 50250 520.6082OK 183.37
24 192.168.100.2 1024 1099857655.37 0.000714 0.772282 50250 508.3352OK 184.39
25 192.168.100.2 1025 1099857656.37 0.000703 0.770671 50250 509.3978OK 185.39
26 192.168.100.2 1026 1099857657.38 0.000658 0.769074 50250 510.4556OK 186.39
27 192.168.100.2 1027 1099857658.38 0.000729 0.737797 50250 532.095OK 187.36
28 192.168.100.2 1028 1099857659.38 0.000663 0.866239 50250 453.1984OK 188.49
29 192.168.100.2 1029 1099857660.38 0.000665 0.764224 50250 513.6951OK 189.39
30 192.168.100.2 1030 1099857661.38 8.329544 106.42267 50250 3.688858OK 296.05
31 192.168.100.2 1031 1099857767.81 8.372031 66.668615 50250 5.8885OK 362.72
32 192.168.100.2 1032 1099857834.48 0.000801 0.769722 50250 510.0259OK 363.49
33 192.168.100.2 1033 1099857835.48 0.000722 0.768225 50250 511.0197OK 364.49
34 192.168.100.2 1034 1099857836.48 0.00067 0.766602 50250 512.1016OK 365.49
35 192.168.100.2 1035 1099857837.48 0.000696 0.765006 50250 513.17OK 366.49
36 192.168.100.2 1036 1099857838.48 0.000681 0.753576 50250 520.9536OK 367.48
37 192.168.100.2 1037 1099857839.48 0.000643 0.872173 50250 450.115OK 368.60
38 192.168.100.2 1038 1099857840.49 0.000695 0.77016 50250 509.7358OK 369.50
39 192.168.100.2 1039 1099857841.49 0.000727 0.768564 50250 510.7943OK 370.50
40 192.168.100.2 1040 1099857842.49 0.000636 0.766913 50250 511.894OK 371.50
0.000692 0.7697
17 192.168.100.3 1017 1099857538.92
8 192.168.100.3 1018 1099857539.93 0.000881 0.73859 5025
0 192.168.100.3 1020 1099857541.93 0
7 192.168.100.1 1057 1099857538.82
57