Measuring SCTP Throughput and Jitter over Heterogeneous Networks

0 downloads 0 Views 208KB Size Report
two Linux laptops and one wireless access point (see Table II). The laptops were ... PSϵ{32 64 128 256 512 1024 1460}Byte PRϵ{100 1000 10000}pps w2w-ap.
Measuring SCTP Throughput and Jitter over Heterogeneous Networks Donato Emma, Salvatore Loreto, Antonio Pescap´e, and Giorgio Ventre Universit`a di Napoli “Federico II”, Naples (Italy) Email: {doemma,pescape,giorgio}@unina.it, [email protected]

Abstract— Stream Control Transmission Protocol (SCTP) is gaining ever more attention. Several proposals, based on simulation results, aiming to replace TCP with SCTP are present. To the best of our knowledge there are no available tools supporting SCTP traffic generation. In this work, to study the performance of SCTP in real heterogeneous (wired/wireless) scenarios, we provide some preliminary results in terms of throughput and jitter that we obtained using a tool we developed to support SCTP traffic generation in real networks. A comparison with TCP and UDP over the same scenarios is also present.

I. I NTRODUCTION The quality of Internet Applications perceived by final users depends on many factors. Among the more important ones there are delay, jitter, and throughput. Understanding how these parameters are influenced by network conditions, by used protocols, and by traffic profiles may prove to be useful in several cases: (i) in the definition and implementation of Service Level Agreements (SLAs); (ii) in the design of new network applications; (iii) in building new network infrastructures; (iv) in the design of new QoS aware routing algorithms; etc. In the last few years several research efforts have been devoted to study the performance of both TCP and UDP over wired, wireless, and heterogeneous network scenarios. Recently, the Stream Control Transmission Protocol (SCTP) [15] and the Datagram Congestion Control Protocol (DCCP) [10] have attracted the attention of the research and industry communities. SCTP was developed to support signaling in Voice Over IP like applications. Considering that many applications may gain advantages from its peculiarities, SCTP is now considered as a general purpose transport protocol. Therefore, there is an increasing interest about the performance of SCTP over real networks. DCCP is a fairly new unreliable datagrams transport protocol that provides a congestion-controlled flow. DCCP should make easy to deploy delay-sensitive application, without risking congestion collapse. In this paper we propose an experimental analysis of the performance of SCTP in several realistic wired, wireless, and heterogeneous network scenarios. This analysis is based on two well known QoS parameters like throughput and jitter. We also propose a comparison among SCTP, UDP and TCP in all the scenarios we investigated. The rest of this paper is organized as follows. Section II presents related works and motivations of our work. In Section III SCTP is introduced. In Section IV we present both the experimental setup used for our analysis and results related

to the considered QoS parameters. Finally, Section V ends the paper with some remarks and issues for future research. Due to space limitation of the camera ready version Tables containing the complete results of our experimental analysis are present at [1]. II. M OTIVATIONS AND R ELATED W ORKS As for the study of SCTP performance, a lot of works carried out in simulation, or analytical analyzed are presented. Also, there is no evidence of studies focused on heterogeneous scenarios. In the case of the wired scenario, authors in [13] presented results related to the experimentation of SCTP over high speed wide area networks. In [14] the authors compare the performance of SCTP and TCP with respect to Web traffic. In [9], using ns-2, the authors study the multi-streaming and the multi-homing SCTP features. They prove that these features have advantages over TCP in the given scenarios. In particular, they define the optimal number of streams in multi-streaming and explain how it affects network performance. In the case of wireless networks, in [12] the authors developed an analytical model that takes into account the congestion window, the round trip time, the slow start and congestion avoidance processes to predict the SCTP performance. By comparing numerical results from the analytical model with simulation results, they demonstrate that the proposed model is able to accurately predict SCTP throughput. In [8] the authors presented a simulation study of delay spike of SCTP, TCP Reno, and Eifel over wirelss links. They found that Eifel performs better than TCP Reno and SCTP when there are no packet losses. However, the opposite is happen when packets are lost in the presence of delay spikes. Also they showed that a higher link bandwidth does not always increase the data throughput of SCTP, TCP Reno, and Eifel. In [11] the authors provide a simulation-based performance comparison of SCTP vs TCP in MANET environments. They found that SCTP and TCP have similar behaviour in MANETs environment, but TCP outperforms SCTP in most cases because of the extra overhead present in SCTP. In [4] the authors presented their simulation results regarding the performance of SCTP in a wireless ad-hoc network environment. Finally, in [7] the authors introduce the main features of SCTP and discuss the state of the art in SCTP research and development activities. They also provide an useful survey of the available products that use SCTP. The lack of networking tools supporting SCTP was witnessed also by this work. Stepping from the screening

Proceedings of the 20th International Conference on Advanced Information Networking and Applications (AINA’06) 1550-445X/06 $20.00 © 2006

IEEE

of the literature we found a lack of experimental studies focusing on the performance evaluation of real scenarios where a comparison among SCTP, TCP, and UDP is presented. Our work extends the results in literature in that (i) we introduce a tool supporting SCTP traffic generation; (ii) we provide a complete evaluation of a wide range of heterogeneous network scenarios in terms of throughput and jitter; (iii) by applying a per-packet analysis we compare SCTP with TCP and UDP. III. SCTP: A BRIEF OVERVIEW The Stream Control Transmission Protocol (SCTP) was defined by the IETF in the late 2000 [15]. It is a session oriented transport protocol that, like TCP, provides a reliable service. Moreover, it provides a lot of new functionalities with respect to TCP and UDP. These new functionalities are critical for signaling in telephony application, and may be also useful for other kinds of applications. Among the peculiar features of SCTP there are the following: (i) Multi Streaming, data transmitted over SCTP can be partitioned into multiple streams. The delivery sequence of these streams is managed independently. Therefore, Packets loss in one stream does not affect the remaining streams. When this feature of SCTP is used the in-sequence delivery of the transmitted data can be ensured only within each stream, not within the whole association. (ii) Multi Homing SCTP supports nodes multihoming. Indeed, any endpoint of an SCTP association may use multiple IP addresses. These addresses are used to cope with physical network failures and other problems of that kind. The SCTP packet format contains two main parts, common header and one or more chunks. The common header contains: (i) source and a destination port numbers; (ii) a 32bit tag used to avoid the insertion of out-of-date or false messages; (iii) a 32bit checksum. The chunks included in the message may be control chunks or data chunks. Each chunk contains: (a) an indication of its type (data or control); (b) a flag that in data chunks is used to control the segmentation and reassembly process, whereas in control chunks may assume different meanings; (c) chunk length; (d) a chunk value. In data chunks the value field contains: (i) a TSN that is a unique sequence number within the association; (ii) a Stream Identifier that identifies the stream to which the following user data belongs; (iii) a Stream Sequence number that represents the stream sequence number of the following user data within the stream; (iv) a Payload Protocol Identifier that represents the upper layer specified protocol identifier; (v) a User Data tath is the payload user data. IV. E XPERIMENTAL A NALYSIS In this section we explore the performance of SCTP with respect to jitter and throughput, considering several network scenarios, with different traffic loads. Moreover, we provide a comparative analysis between SCTP and both TCP and UDP. A. Experimental Setup Network Scenarios: In exploring the performance of SCTP we studied its behavior in a lot of heterogeneous

network scenarios (see Table I). In all these scenarios we used two hosts: the first one acting as a traffic source; the second one acting as a traffic sink. In order to investigate the characteristics of the current implementation of SCTP under “quasi-ideal” conditions, we first connected the two hosts via an Ethernet crossover cable (eth2eth). eth2eth offers the higher throughput, the lower loss rate, and the lower delay among all the scenarios we studied. Then we evaluated SCTP in two wired to wireless scenarios. In these scenarios one host is connected via an Ethernet cable to an access point (AP in the following), whereas the other is connected to the AP via an 802.11 connection. We first considered the “wired host” acting as a traffic source and the “wireless host” acting as a traffic sink (e2w). Then we exchanged the roles of the hosts using the wireless one as a traffic source and the wired one as a traffic sink (w2e). Finally, we considered two wireless scenarios. In the first one the two hosts communicate via AP (w2w-ap). In the second one the two hosts are connected directly in “ad-hoc mode” (w2w-ah). As for the traffic, we used an “active approach”. More precisely, in order to limit the number of variable parameters that must be taken into account in our analysis, we generated constant traffic sources (that is constant packet rate (PR) and constant payload size (PS)) using UDP, TCP and SCTP. In all the scenarios, except eth2eth, we considered three working conditions: low, medium, and high packet rate. In the first one we sent 100 packets per second. In the second one we sent 1000 packets per second. In the third one we sent 10000 packets per second. As for the packet size, we used a payload size ranging from about 32B to 1460B. As mentioned above, we considered eth2eth in order to evaluate the behavior of SCTP in a scenario that is characterized by high available bandwidth, low delay, and low jitter. Therefore, in this scenario we decided to fix the PS to 512Byte and we varied the packet rate in order to obtain different load conditions ranging from about 40M bps to about 95M bps. Test-bed description: We used a test-bed composed of two Linux laptops and one wireless access point (see Table II). The laptops were involved in all the scenarios we considered. More precisely augello was always used as a traffic source, and fazio was always used as a traffic sink. The wireless AP is used only in the e2w, w2e, and w2w-ap scenarios. Software Tool: We used D-ITG [5] to generate traffic flows in our experimental analysis. D-ITG is a packet level traffic generator we have developed in the last few years. The current publicly available release is the 2.4. This release does not support SCTP traffic generation. Therefore we extended D-ITG in order to support SCTP. D-ITG extension is based on the LK-SCTP [2] implementation of SCTP that is included in the Linux 2.6 Kernel family. D-ITG can be used to analyze throughput and jitter at packet level, and also to measure the round trip time delay and the one way delay. In our analysis, we have not considered the delay. Indeed, the delay that characterize some of the scenarios we investigated (i.e. the eth2eth) is too small to be correctly studied without the support of a special hardware to synchronize the components of the

Proceedings of the 20th International Conference on Advanced Information Networking and Applications (AINA’06) 1550-445X/06 $20.00 © 2006

IEEE

TABLE I D ESCRIPTION OF THE NETWORK SCENARIOS Scenario Name

Scenario Description

eth2eth e2w

Two host connected via an Ethernet crossover cable Two host that communicate via an access point (AP). The first one is connected to the AP via an Ethernet cable. The second one is wireless connected to the AP. The first host acts like a traffic source Similar to the e2w scenario. In this case the wireless connected host acts like a traffic source Two wireless hosts connected via an AP Two wireless hosts connected in “ad hoc” mode

w2e w2w-ap w2w-ah

Medium Transmission Rate 100Mbps 11Mbps

Traffic Profile

11Mbps

P S{32 64 128 256 512 1024 1460}Byte P R{100 1000 10000}pps

11Mbps 11Mbps

P S{32 64 128 256 512 1024 1460}Byte P R{100 1000 10000}pps P S{32 64 128 256 512 1024 1460}Byte P R{100 1000 10000}pps

P S = 512Byte P R{9766 12207 14648 17090 19531 20752}pps P S{32 64 128 256 512 1024 1460}Byte P R{100 1000 10000}pps

TABLE II T EST- BED DESCRIPTION Host Name Augello Fazio

Montalbano

Description Asus V6800V CPU Intel Pentium M processor 2.00GHz RAM size 2GB Toshiba Satellite S5200-801 CPU Intel Pentium 4 processor 2.00GHz RAM size 512MB Wireless Broadband Internet Gateway with Built in 4-Port Switch (DLINK DI 624+)

O.S. version Linux Kernel 2.6.10-5 Linux Kernel 2.6.10-5

Firmware version 1.19b09

test-bed, and to capture the traffic with a sufficient precision. Anyway, throughput and jitter are the main QoS parameters that should be considered in delivering many services [3]. B. Experimental Results In this section we report and discuss results of our experimental analysis. In this experiment we do not use neither multistream neither multihoming SCTP’s futures. All the traffic between the two SCTP endpoints is delivered only on the stream 0 in an ordered way. SCTP adopts the same TCP congestion/flow control scheme, with the exception of the fast recovery mechanism, therefore we expect a similar behavior between TCP and SCTP also if an inefficiency in the current specification of SCTP’s congestion control degrades the performance when there are multiple packet losses in a single window [6]. Moreover in the analysis of the experimental results it is necessary to take into consideration the well known poor performance and efficiency of LK-SCTP implementation. Before to step into experimental details, it is worth mentioning that we repeated each test several times. In the following graphics and tables the mean values across 20 test repetitions are reported. Figure 1 shows throughput and jitter we measured in the eth2eth scenario. In this scenario SCTP performances are comparable to that of UDP and TCP. Both UDP and SCTP are capable to serve a bitrate of 85M bps. TCP seems to have a better control of the congestion of the link. It can serve a maximum bitrate of about 90M bps. As for the jitter the three protocols are characterized by a similar behavior. There is an initial decreasing in the values we measured until the communication becomes congested. During the link congestion we experimented a sudden increase of the jitter. UDP is the protocol that presents the minimum jitter in all the non congested traffic conditions. Vice versa, TCP presents the highest jitter when the link is not congested. Finally

Ethernet Card Marvell Technology Group Ltd. Yukon Gigabit Ethernet Intel Corp. 82801CAM (ICH3) PRO/100 VE Ethernet Controller

Wireless Card Intel Corp. PRO/Wireless 2200BG

built-in wireless LAN connection and 10/100Mbps switch

Wireless LAN standards supported: 802.11b, 802.11b+, 802.11g

SCTP presents intermediate jitter values. Tables I, II, III, and IV (contained in the PAEMN06Tables.pdf document at [1]) contain throughput and jitter values, which we measured in wireless and heterogeneous scenarios. More precisely in these tables we reported mean, median, and standard deviations values as functions of the packet size for all the working conditions we considered. In the following we discuss values we measured in each working condition. Low packet rate: In all the network scenarios SCTP has been able to follow the growing of the throughput we imposed by increasing the PS from 32Byte to 1460Byte. More precisely, it reached a maximum throughput equal to 1.17e+3 Kbps for P S = 1460Byte. As for the jitter, excluding the w2w-ah scenario, we measured: (i) mean values that are less than 1ms; (ii) standard deviations and median values that are comparable with jitter mean values. Therefore, we found a high dispersion of the jitter samples around their mean values. However, the maximum jitter values that we measured are always in the order of magnitude of few fractions of milliseconds. Therefore we may argue that such dispersion should have a little impact on SCTP based applications that require stringent jitter constraints. In the w2w-ah scenario there is a quite different behavior. Both the mean and the maximum jitter values we measured are higher then those we measured in the other scenarios. The maximum jitter we measured is equal to about 6.8ms. One order of magnitude higher than that of the other scenarios. Both UDP and TCP showed a throughput behavior similar to that of SCTP. Indeed they have been able to correctly serve all the transmission rates we imposed. As for the jitter, the behavior of UDP is very similar to that of SCTP. Also the TCP jitter behavior, excluding a singularity we found for P S = 1024Byte, can be assimilated to that of SCTP. In all the scenarios we considered, with a payload size equal to 1024Byte TCP showed a higher

Proceedings of the 20th International Conference on Advanced Information Networking and Applications (AINA’06) 1550-445X/06 $20.00 © 2006

IEEE

Lucent Orinoco GOLD Wireless PC Card - 11Mbit/s

100

80 70 60 50

50

0.09

60 70 80 sent bitrate [Mbps]

90

60 50

100

UDP jitter

0.08 0.07 0.06

jitter [ms]

jitter [ms]

70

40 40

0.05 0.04 0.03 0.02 0.01 0

80

40

50

60 70 80 sent bitrate [Mbps]

Fig. 1.

90

100

0.065 0.06 0.055 0.05 0.045 0.04 0.035 0.03 0.025 0.02 0.015

TCP Throughput

90 80 70 60 50 40

40

50

60 70 80 sent bitrate [Mbps]

90

100

40

50

60 70 80 sent bitrate [Mbps]

0.09

SCTP jitter

0.06 0.05 0.04 0.03

40

50

60 70 80 sent bitrate [Mbps]

90

100

0.02

40

50

60 70 80 sent bitrate [Mbps]

90

100

Throughput and jitter measured in the eth2eth scenario

Medium packet rate: In all the network scenarios that we considered SCTP has not been able to reach the maximum throughput that we asked for. There is a threshold starting from which the measured throughput was less than the requested one. This threshold depends on the network scenario. For example, in e2w SCTP has been capable to follow the requested throughput until 2.05e+3 Kbps. Starting from this value there is an increasing difference between the throughput we tried to impose and the measured one. However, SCTP has reached a maximum throughput equal to 4.14e+3 Kbps for P S = 1024Byte (for this PS the throughput we tried to impose was equal 8.19e+3 Kbps). As for the jitter, excluding the w2w-ah scenario and considering only the cases in which SCTP has been able to serve the required bitrate, we have measured mean values always less than 2ms and standard deviations at least one order of magnitude less than the mean values. Therefore, in this working condition, considering only the PSs for which the SCTP association was not saturated, the jitter is quite small and it is characterized by a regular behavior. Moreover, jitter mean values do not dramatically grow when the SCTP channel became overloaded. Also in the medium packet rate tests, in the w2w-ah scenario the mean and the maximum values we measured are one order of magnitude higher than those of the other scenarios. As for the throughput, excluding the w2w-ap scenario, for UDP we measured throughput values very similar to that of SCTP. Indeed, we have found for both UDP a SCTP the same throughput-thresholds that is always less than that of TCP. In the w2w-ap scenario for both TCP and SCTP we observed a behaviour that is very similar to that of the other scenarios. UDP, instead, has been unable to reach the imposed throughput also for the minimum PS. Indeed, for P S = 32Byte we measured a throughput of 1.82e+2 Kbps when the expected one was of 2.56e+2 Kbps. As for the jitter, if we consider only the cases in which the SCTP connection was not satured, for

TCP we measured mean jitter values always comparable with that of SCTP; for UDP we measured jitter values comparable with that of SCTP in the w2w-ap scenario and quite smaller in the others. High packet rate: In order to not exceed the nominal throughput of the wireless channels we have considered only three packet sizes: 32, 64, and 128 Bytes. In all the scenarios, SCTP has not been able to reach the requested throughput. As for the jitter, we measured mean values that range from about 1e-1ms to about 1.5e+1 ms with standard deviations that are comparable or one order of magnitude smaller than the mean values. UDP showed a behavior very similar to that of SCTP. Vice versa, TCP achieved better performance. It followed the required throughput for P S = 32Byte in all the network scenarios, and for P S = 64Byte in the w2e scenario. Also its jitter was less than those of both UDP and SCTP. Summary of the results in the heterogeneos scenarios: In this paragraph we highlight some results we obtained in the heterogeneous scenarios. Figure 2 shows throughput and jitter as a function of the time in the e2w and w2e scenarios with P S = 512Byte and P R = 1000pkt/s. In both these scenarios we found that the throughput of SCTP is less than those of both TCP and UDP. As for the jitter, SCTP presents the worst values in the e2w scenario, and a jitter that is comparable with that of TCP in the w2e scenario. Moreover, if we consider a higher packet rate the behavior of SCTP is characterized by a greater variability. Indeed, Figure 3 shows the throughput and the jitter we measured in the e2w scenario with P S = 64Byte and P R = 10000pkt/s. We found a “quasi periodic” trend of both throughput and jitter. A deeper analysis is needed in order to identify the causes of this periodical behavior. Therefore, we are induced - from these first results - to assume that there are no advantages if SCTP is used as a simple substitute of TCP and UDP. As for a comparison between the heterogeneous and the wireless scenarios we found some minor differences in the behavior of SCTP. In both e2w and w2e, with a P R = 1000pkt/s, SCTP has been able to support a maximum bitrate equal to 2.05e+3 Kbps. This value is higher than that SCTP reached in the w2w-ap scenario (1.02e + 3Kbps) and equal to that of

Proceedings of the 20th International Conference on Advanced Information Networking and Applications (AINA’06)

IEEE

100

0.07

jitter, equal to about 5.5ms. This singularity requires a deeper analysis, but we argue that, for this particular combination of packet size and packet rate, the TCP jitter is mainly determined by the TCP buffer management overhead, with a little dependence on the channel state.

1550-445X/06 $20.00 © 2006

90

TCP jitter

0.08

jitter [ms]

40

100

SCTP Throughput

90

received bitrate [Mbps]

UDP Throughput

90

received bitrate [Mbps]

received bitrate [Mbps]

100

2

UDP Throughput Throughput 4200 PS = 512Byte; PR = 1000pkt/s; e2w Scenario SCTP TCP Throughput

UDP Jitter SCTP Jitter TCP Jitter

PS = 512Byte; PR = 1000pkt/s; e2w Scenario

4000

Jitter [ms]

throughput [Kbps]

1.5

3800

3600

1

0.5

3400 10

20

30

40

50

60

70

0

80

10

20

30

40

70

90

UDP Jitter SCTP Jitter TCP Jitter

PS = 512Byte; PR = 1000pkt/s; w2e Scenario

2.2 2

4000

jitter [ms]

throughput [Kbps]

60

2.4

UDP Throughput 4200 PS = 512Byte; PR = 1000pkt/s; w2e Scenario SCTP Throughput TCP Throughput 4100

3900 3800

1.6

1.2

3600 3500

1.8

1.4

3700

10

20

30

40

50

60

70

1

80

10

20

30

40

50

60

70

80

time [s]

time [s]

Fig. 2.

50 time [s]

time [s]

Throughput and jitter of the e2w and w2e scenarios (P S = 512Byte; P R = 1000pkt/s)

5000

25

UDP Throughput PS = 64Byte; PR = 10000pkt/s; e2w Scenario SCTP Throughput TCP Throughput

PS = 64Byte; PR = 10000pkt/s; e2w Scenario

UDP Jitter SCTP Jitter TCP Jitter

4000

jitter [ms]

throughput [Kbps]

20

3000

10

2000

5

1000 0

10

20

30

40

50

60

70

80

time [s]

Fig. 3.

15

0

0

10

20

30

40

50

60

70

time [s]

Throughput and jitter of the e2w scenario (P S = 64Byte; P R = 10000pkt/s)

the w2w-ah scenario. Moreover, with P R = 10000pkt/s the maximum throughput reached by SCTP in the heterogeneous scenarios was greater than that it reached in the two wireless scenarios. V. C ONCLUSION In this paper we presented a packet level experimental analysis of SCTP in wired, wireless and heterogeneous (wired/wireless) scenarios in terms of throughput and jitter. We also compared SCTP with TCP and UDP. We introduced a traffic generation tool that supports UDP, TCP and SCTP traffic generation at packet level. Results presented in this paper may define a reference framework for the development of throughput and jitter aware SCTP based network applications. Moreover, we found some singularities and peculiar behaviors in our measurements that require a deeper analysis. Therefore, we have planned to: (i) apply the presented methodology to other QoS parameters like packet loss and delay; (ii) complete the statistical characterization of SCTP in heterogeneous scenarios; (iii) perform a comprehensive analysis that is aimed at explaining the singularities that we found. ACKNOWLEDGMENT This work has been partially supported by the MIUR in the framework of both the PRIN 2004 QUASAR project and the Webminds FIRB project. R EFERENCES

[3] M. Alves, L. Corsello, D. Karrenberg, and C. Ogut. New measurements with the ripe ncc test traffic measurements setup. In Proc. of Passive and Active Measurement Conference (PAM), ’02. [4] A. Argyriou and V. Madisetti. Performance evaluation and optimization of sctp in wireless ad-hoc networks. In Proc. of the 28th Annual IEEE International Conference on Local Computer Networks (LCN’03), ’03. [5] S. Avallone, D. Emma, and A. Pescap´e. D-itg, distributed internet traffic generator web site: http://www.grid.unina.it/software/itg/, as of Sept. ’05. [6] A. L. Caro, K. Shah, J. R. Iyengar, and P. D. Amer. Sctp and tcp variants: Congestion control under multiple losses. Technical Report 2003-04, CISC Dept, Univ of Delaware, ’03. [7] S. Fu and M. Atiquzzaman. Sctp: state of the art in research, products, and technical challenges. IEEE Communications Magazine, 42(4):64 – 76, Apr. ’04. [8] W. Ivancic, S. Fu, and M. Atiquzzaman. Effect of delay spike on sctp, tcp reno, and eifel in a wireless mobile environment, Oct. ’02. [9] S. Kang and M. Fields. Experimental study of the sctp compared to tcp. Technical report, Department of Electrical Engineering of Texas A & M University, ’03. [10] E. Kohler, M. Handley, and S. Floyd. Datagram congestion control protocol (dccp). IETF Internet Draft, draft-ietf-dccp-spec-11.txt. [11] A. Kumar, L. Jacob, and A. L. Ananda. Sctp vs tcp: Performance comparison in manets. In Proc. of the 29th Annual IEEE International Conference on Local Computer Networks (LCN’04), Washington, DC, USA, ’04. IEEE Computer Society. [12] L. Ma, F. Yu, and V. C. M. Leung. Modeling sctp throughput in integrated wlan/cellular networks, May ’05. [13] D. Nagamalai and J.-K. Lee. Performance of sctp over high speed wide area networks, Dec. ’04. [14] R. Rajamani, S. Kumar, and N. Gupta. Sctp versus tcp: Comparing the performance of transport protocols for web trafc. Technical report, Computer Sciences Department of University of Wisconsin Madison, July ’02. [15] R. Stewart, Q. Xie, K. Morneault, C. Sharp, H. Schwarzbauer, T. Taylor, I. Rytina, M. Kalla, L. Zhang, and V. Paxson. Stream control transmission protocol. RFC 2960, Oct. ’00.

[1] www.grid.unina.it/software/ITG/D-ITGpublications/PAEMN06Tables.pdf. [2] Linux kernel stream control transmission protocol (lksctp) web site: http://lksctp.sourceforge.net/, as of Dec. ’05.

Proceedings of the 20th International Conference on Advanced Information Networking and Applications (AINA’06) 1550-445X/06 $20.00 © 2006

IEEE