Elsevier

Computer Networks

Volume 50, Issue 13, 15 September 2006, Pages 2160-2182
Computer Networks

Throughput models for SCTP with parallel subflows,☆☆

https://doi.org/10.1016/j.comnet.2005.08.001Get rights and content

Abstract

Multi-streaming provides SCTP with an aggregation mechanism to accommodate heterogeneous objects, which belong to the same application but may require different type of QoS from the network. However, in the current SCTP specification, supporting QoS inside an SCTP association is not a concern; it is also hindered by the fact that all streams share the flow and congestion control at the association level. We introduce the concept of subflow and propose to modify the current SCTP such that the streams are grouped into several subflows according to their required QoS. To prevent false sharing, each subflow in the modified SCTP (SF-SCTP) implements its own flow and congestion control. Our design also integrates fractional congestion control to improve the fairness of the modified SCTP towards other network traffic. Analytic models are presented to compare the throughput of the original SCTP and SF-SCTP using different flow and congestion control mechanisms. Simulations with ns-2 are used to estimate the throughput in a simplified Diff-Serv network. The analytic models are confirmed to accurately reflect the SCTP behavior. The simulation shows that our proposed solution is able to support QoS among the SCTP streams, and that it achieves a better throughput. The results also indicate that SF-SCTP is more fair to the original SCTP after fractional congestion control is introduced.

Introduction

The Stream Control Transmission Protocol (SCTP) (RFC 2960) [9], [27] is a new transport layer protocol designed to overcome the shortcomings and limitations of the UDP and TCP. With its new services such as multi-homing, multi-streaming, message boundary preservation, and enhanced security features, SCTP is more suitable for applications requiring high performance and reliability.

Originally, SCTP includes the multi-streaming feature to decouple the reliable transfer of user data from the strict ordered delivery. SCTP’s support for multiple, logically independent message streams is an example of partial ordered delivery, which is extremely useful for application with strict timing requirement [7], [21] by alleviating the head-of-line blocking problem. This feature reduces buffer requirements at the receiver, making SCTP an attractive transport protocol for resource limited wireless handheld devices [5]. Multi-streaming enables SCTP to multiplex related, yet independent data streams over a single end-to-end association [26]. With multi-streaming, SCTP is equipped with an internal mechanism to support concurrent transmission of multiple objects [19]. For example, HTTP using SCTP can load a web page with multiple objects by a single SCTP association instead of several TCP connections.

Although multi-streaming lets SCTP accommodate heterogenous objects with different Quality of Service (QoS), the current SCTP does not take advantage of this feature. It is unaware of QoS and lacks an internal mechanism to support preferential treatment of SCTP streams. First, the SCTP association may find it hard to mark an outgoing packet’s QoS byte since it could encapsulate messages from streams with different QoS requirements. Second, the current SCTP employs a TCP-like congestion control mechanism at the association level, which means that the streams carrying different objects are treated according to the same congestion state information. This shared congestion information could help to improve the overall SCTP performance if the SCTP packets are treated the same by the network [6]. However, when packets with different QoS markings are processed differently by the network (in terms of packet loss rate, delay, etc.), the shared congestion information could result in the so-called false sharing [3]. This phenomenon occurs when flows sharing the congestion information do not actually share the same network bottleneck. The occurrence of false sharing could worsen the overall performance of SCTP association. In the case where streams have different packet drop rates, the transport layer tries to moderate the performance for all streams; hence, the performance of the higher priority streams will be penalized due to the shared information from the lower priority streams. When streams experience different round trip times (RTTs), which may lead to unnecessary Fast Retransmits and time-outs, false sharing severely affects the SCTP performance in that all streams are penalized, regardless of their priorities.

In this paper, we define the concept of subflow [25], [33], [34] and present the necessary modifications to the SCTP specification to support preferential treatment of SCTP streams. In this modified SCTP (called SF-SCTP), each subflow has its own flow and congestion control and consists of SCTP streams that require the same type of QoS from the network. Depending on the QoS required, an SCTP association may have one or more subflows, which serve as the independent transmission channels between an SCTP communication peer. Since the congestion state information is only shared by the SCTP streams requesting the same QoS, our design inherently avoids false sharing.

Although SF-SCTP has multiple subflows encapsulated, the SF-SCTP design in this paper is targeted to serve a single application with different QoS requirements for its data. Different SCTP applications will each use a single association, which may be either original SCTP or SF-SCTP. The inclusion of multiple subflows in one SCTP association effectively eliminates the need for an SCTP application to open and maintain multiple connections when it needs to transmit several objects concurrently. In addition, using SF-SCTP, the application can now give different priorities to the objects in the same communication session according to their relative importance. Since the design of SF-SCTP takes advantage of the multi-streaming feature of SCTP, it only adds a thin layer into the existing SCTP implementation and is backward compatible.

Due to multiple subflows, SF-SCTP behavior is similar to the aggregation of concurrent original SCTP associations. As a result, SF-SCTP has a much faster growing congestion window and is more resistant to packet losses, which makes SF-SCTP more aggressive than the original SCTP. As the number of subflows increases, the SF-SCTP behavior gets more aggressive, thereby having a negative impact of increasing packet drop rates in the network [11]. Our SF-SCTP may thus perform better in terms of its throughput by unfairly harming other traffic on the network. To alleviate the over-aggressiveness of SF-SCTP, we introduce to SCTP a mechanism called fractional congestion control (FCC), which was originally suggested for TCP [14], [15]. FCC reduces the aggressiveness of SF-SCTP by increasing the subflow congestion window slower than the original SCTP in congestion avoidance phase.

Starting from the existing TCP throughput models [23], we derive throughput models for the original SCTP and several different implementations of SF-SCTP. The simulation experiments on ns-2 demonstrate that our SF-SCTP design is able to support preferential treatment of SCTP streams and achieve higher throughput than the original SCTP by taking advantage of the QoS provided by the network. The simulations also point out that, for a network where all packet losses are due to congestion, SF-SCTP with FCC reduces its aggressiveness but still performs somewhat unfairly towards the original SCTP. This unfairness stems from the fact that SF-SCTP with FCC experiences a smaller packet drop rate than the original SCTP in such a network.

Our work contributes to SCTP research in several ways. We present a novel design that combines the concept of subflows (introduced in Refs. [25], [33], [34]) and fractional congestion control (introduced for TCP in Refs. [14], [15]). To the best of our knowledge, this paper is the first to introduce subflow-capable SCTP with FCC. In addition, we implemented SF-SCTP with FCC in NS-2 platform to perform extensive simulation experiments and provide insights into performance gains for applications with diverse QoS requirements. We extended the research of Refs. [14], [15] to study the effect of FCC on the throughput of SF-SCTP with different congestion window growth rate settings. We find that FCC, while effective in reducing SF-SCTP’s aggressiveness, can also reduce its packet drop rate, a new phenomenon not addressed in Refs. [14], [15].

Section 2 outlines the techniques to improve the transport layer performance. The SF-SCTP design is in Section 3. Analytic models for different implementations of SF-SCTP are introduced in Section 4. The results of ns-2 experiments are in Section 5. Section 6 presents conclusions and guidelines for future work.

Section snippets

Related work

Several techniques to improve the performance of transport layer have been proposed [3], [6], [11]. In Ref. [6], the authors present an end-system architecture centered around a Congestion Manager (CM) that ensures proper congestion behavior and allows applications to easily adapt to network congestion. The CM is a middle layer responsible for congestion control of all TCP connections between TCP and IP layers. Independent multiple TCP connections cooperate rather than compete with each other.

Motivation and applications

Multi-streaming can be particularly useful for applications that need to exchange multiple, logically separate message streams between two endpoints (RFC 3257). It has been shown to improve FTP-ing multiple files, e.g., for system backups or mirror-site downloads. Multi-streaming is also useful for applications that multiplex related, yet independent data streams (e.g., voice, text, video) subject to different QoS requirements [9]. Ref. [24] recommends mapping SIP (RFC 3261) transactions into

Analytic models

In this section, we derive SCTP’s analytic models in a simplified DS network where each DSCP is mapped to a unique packet drop rate. In this DS network, the flows with different DSCPs shall experience different packet drop rates after reaching steady state since the network treats each flow with different priorities. It is also assumed in this simplified DS network that all the flows between a sender and receiver shall travel through the same physical path so that they will have the same RTT.

Simulation experiments and analytical throughput

In this section, we report the results of the simulation experiments conducted using ns-2 [29] with the integrated SCTP extensions from University of Delaware [8]. Two simulation scenarios are created with the topologies shown in Fig. 7, Fig. 10, respectively. The obtained results from Scenario I demonstrate SF-SCTP’s ability to support QoS and also, without FCC, SF-SCTP’s over-aggressiveness compared to the original SCTP and SF-SCTP-shared. The results from the Scenario II point out that

Concluding remarks

With multi-streaming, SCTP was equipped with the capability to concurrently transmit messages with different QoS requirements. However, when SCTP was originally introduced, supporting QoS inside an SCTP association was not addressed. In this paper, we propose a modification to the current SCTP specification to support preferential treatment of SCTP streams.

We define the necessary modifications to the current SCTP to make it subflow-capable: namely, a new data chunk and SACK format to implement

Jianping Zou received his B.S. degree in Electrical Engineering from Zhejiang University, Hangzhou, China and his M.S. in Electrical and Computer Engineering from Polytechnic University, Brooklyn, NY, in 2001 and 2003, respectively. He is currently a Ph.D. candidate in Electrical Engineering Department at the Graduate Center of the City University of New York. His research interests are in the field of computer and communication networks, with emphasis on the design of transport layer protocols

References (34)

  • A. Abd El Al et al.

    Improving throughput and reliability in mobile wireless networks via transport layer bandwidth aggregation

    Comput. Network (Special Issue on Military Communications Systems and Technologies)

    (2004)
  • A. Abd El Al et al.

    LS-SCTP: A bandwidth aggregation technique for stream control transmission protocol

    Comput. Commun. (Special Issue on Protocol Engineering for Wired and Wireless Networks)

    (2004)
  • A. Akella, H. Balakrishnan, S. Seshan, The impact of false sharing on shared congestion management, in: Proc. IEEE Int....
  • P. Almquist, Type of service in the internet protocol suite. RFC 1349, IETF,...
  • M. Atiquzzaman, W. Ivancic, Evaluation of SCTP multistreaming over satellite links, in: Proc. IEEE Int. Conf. Comput....
  • H. Balakrishnan, H.S. Rahul, S. Seshan, An integrated congestion management architecture for Internet hosts, in: Proc....
  • A. Balk, M. Sigler, M. Gerla, M.Y. Sanadidi, Investigation of MPEG-4 video streaming over SCTP, in: Proc. IIIS World...
  • A.L. Caro, J.R. Iyengar, NS-2 SCTP module, version 3.2, 2002. Available from:...
  • A.L. Caro et al.

    SCTP: A proposed standard for robust Internet data transport

    IEEE Comput.

    (2003)
  • D. Clark et al.

    Explicit allocation of best-effort packet delivery service

    IEEE/ACM Trans. Networks

    (1998)
  • S. Floyd et al.

    Promoting the use of end-to-end congestion control in the Internet

    IEEE/ACM Trans. Networks

    (1999)
  • W. Gropp et al.

    Using MPI

    (1994)
  • T.J. Hacker, B.D. Athey, B.D. Noble, The end-to-end performance effects of parallel TCP sockets on a lossy wide-area...
  • T.J. Hacker, B.D. Noble, B.D. Athey, The effects of systemic packet loss on aggregate TCP flows, in: Proc. ACM/IEEE...
  • T.J. Hacker, B.D. Noble, B.D. Athey, Improving throughput and maintaining fairness using parallel TCP, in: Proc. IEEE...
  • J. Heinanen, F. Baker, W. Weiss, J. Wroclawski, Assured forwarding PHB group. RFC 2597, IETF,...
  • J. Heinanen, R. Guerin, A two-rate three-color marker. RFC 2698, IETF,...
  • Cited by (7)

    • Performance evaluation of subflow capable SCTP

      2006, Computer Communications
      Citation Excerpt :

      In this way, the backward compatibility is provided. The detailed descriptions for the new chunk formats of SF-SCTP can be found in our earlier work [32–34]. The additional overhead introduced by the SF-SCTP consists of two parts.

    • A fluid flow model for SCTP traffic over the Internet

      2012, Proceedings - 2012 International Conference on Emerging Technologies, ICET 2012
    • Modeling and analysis for steady-state throughput of SCTP based drop-tail queue

      2010, Beijing Jiaotong Daxue Xuebao/Journal of Beijing Jiaotong University
    • Extension of SCTP for concurrent multi-path transfer with parallel subflows

      2010, IEEE Wireless Communications and Networking Conference, WCNC
    • DSA-SCTP: An extension of SCTP to reduce web response time

      2009, Proceedings of 2009 IEEE International Conference on Communications Technology and Applications, IEEE ICCTA2009
    View all citing articles on Scopus

    Jianping Zou received his B.S. degree in Electrical Engineering from Zhejiang University, Hangzhou, China and his M.S. in Electrical and Computer Engineering from Polytechnic University, Brooklyn, NY, in 2001 and 2003, respectively. He is currently a Ph.D. candidate in Electrical Engineering Department at the Graduate Center of the City University of New York. His research interests are in the field of computer and communication networks, with emphasis on the design of transport layer protocols and the implementation of mobile ad-hoc networks in a real-time embedded operating system.

    M. Ümit Uyar is currently with the City College and the Graduate Center of the City University of New York. His interests are in the testing and reliability of computer and communication networks and protocols. He was a Co-Principal Investigator for two multi-million dollar grants from U.S. Army Research Labs, awarded to the City University of New York. He was a Distinguished Member of Technical Staff at AT&T Bell Labs until 1993. In Bell Labs, he received a Vice Presidential Quality Award for co-designing software tools, three AT&T Bell Labs Vice Presidential Research Appreciation Awards, and a Best Paper Award in AT&T Electronic Testing Conference. He was granted “Docent” title by the National University Council of Turkey in 1992. He was the co-chair of the 18th IFIP Int’l. Conference on Testing of Communicating Systems (Testcom 2006), 6th Int’l. Conference on Formal Description Techniques (Forte 1993), and the 12th Int’l. Symposium on Protocol Specification, Testing and Verification (PSTV 1992), He co-edited the book titled “Conformance Testing Methodologies and Architectures for OSI Protocols,” published by the IEEE Computer Society Press. He holds three U.S. patents. He has a B.S. degree from Istanbul Teknik Universitesi, and M.S. and Ph.D. degrees from Cornell University, Ithaca, NY, all in Electrical Engineering.

    Mariusz A. Fecko received M.S. degrees in both Electronics and Computer Science from AGH University of Science and Technology, Poland; and M.S. and Ph.D. in Computer and Information Sciences from the University of Delaware. At UD he jointly developed formal testing methodologies for US Army radio-network protocols. In 2000, Mariusz joined Applied Research at Telcordia Technologies, Inc., New Jersey. He improved the quality of telecom clearinghouses such as Exchange Link/WNP through a novel use of XML technologies. Mariusz serves as Principal Investigator in ARL Collaborative Technology Alliance in wireless networks, and has been a key member of the MOSAIC and PILSNER teams, designing communications technologies for wireless on-the-move networks. He received three Telcordia CEO Team Awards for winning new business. He co-chaired the 18th IFIP Int’l Conf. Testing Communicating Systems (TestCom’06).

    Sunil Samtani is currently a Director managing a group of researchers in the area of Wireless Networking and Network Management. Sunil received B.E. degree in Computer Engineering from University of Bombay; and M.S. in Computer Science from the University of Missouri. He also has an MBA in Finance from the Stern School of Business. Sunil currently serves as the Principal Investigator for the CERDEC PILSNER Program where he leads a team of researchers and engineers from Telcordia, GDC4S, Harris, and BAE to provide optimal network design and directional networking capabilities for WIN-T. In the past, Sunil was PI for DARPA’s Software for Distributed Robotics Program. He is also the lead technical architect of an ad hoc mobility protocol suite for the U.S. Army CERDEC MOSAIC program, and has been the Chief Engineer on the FCS Network Management System effort. He has led a team of researchers to bring CECOM MOSAIC technologies to TRL-6 readiness and also to transition them to key Army acquisition programs such as WIN-T and FCS. His research interests include networking communications services and network and management for mobile ad hoc networks, policy-based management systems, and databases.

    Prepared through collaborative participation in the Communications & Networks Consortium sponsored by the US Army Research Lab under the Collaborative Technology Alliance Program, Cooperative Agreement DAAD19-01-2-0011. The US Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation thereon.

    ☆☆

    © 2005 Telcordia Technologies, Inc. and the City University of New York.

    View full text