Elsevier

Computer Communications

Volume 26, Issue 2, 1 February 2003, Pages 79-89
Computer Communications

Implications of proactive datagram caching on TCP performance in wireless/mobile communications

https://doi.org/10.1016/S0140-3664(02)00108-1Get rights and content

Abstract

The operation of TCP in wireless–mobile environments is discussed in this paper. After briefly presenting previous efforts to ameliorate TCP performance in the considered environments, we propose a new mechanism for tackling the slow-down problems caused by handovers. Our mechanism is based on stochastic datagram relocation. Traffic destined to the mobile terminal is tunneled to and cached into adjacent cells according to the output of a path prediction algorithm. To reduce the associated overhead, only percentages of inbound traffic are copied to the cell's neighborhood on the basis of estimated probabilities. The time scheduling for datagram relocation is also taken into account. Simulations of the proposed architecture show substantial performance improvements for TCP traffic.

Introduction

The operation of TCP in wireless/mobile communications has been an important research issue in recent years, owing to the impressive growth experienced in that area of modern telecommunications. Significant contributions such as the one presented in Ref. [1], indicate that the unmodified, standardized operation of TCP is not well aligned with the peculiarities of cellular environments. Terminal movement across cell boundaries leading to handover is misinterpreted by common TCP implementations as a sign of congestion within the fixed network. To handle such congestion, TCP unnecessarily slows down transmission by reducing window sizes, and performing retransmissions if relevant need arises.

A number of efforts have been reported for the resolution of the above-mentioned problems. Notable contributions are the Snoop protocol [2], [3], the split-connection and end-to-end protocol families. Snoop tries to conceal potential problems caused by the peculiarities of the wireless interface. It involves the installation of a specialized module (Snoop Agent) in the base station (BS). The Agent maintains a local cache, monitors the exchange of TCP packets and acknowledgements, and performs local retransmissions as needed. The split-connection approach handles the inconsistencies between the wireless and wired link. The end-to-end approach copes with the retransmission timeouts and the shrink of the congestion window that causes TCP slowdown. A more extensive discussion on TCP enhancements for wireless networks follows.

The Snoop scheme has been used as a basis for the work presented in this paper. Snoop does not cover the cases where the mobile terminal (MT) gets involved in handovers. In such cases, the Snoop cache accumulated in the BS remains unused since the MT's point of attachment to the fixed network changes. In the new BS, no segments pertaining to the network dialogs, disrupted by the handover are cached. Hence, the problems that Snoop tries to resolve are seriously aggravated.

To address the problem described in the previous paragraph we suggest an alternative architecture. Specifically, we adopt the use of a path prediction algorithm and the proactive formulation of datagram caches in adjacent BSs. This scheme is quite similar to the Daedalus project approach, reported in Ref. [6], where traffic destined to the MT is multicast (by the Home Agent of Mobile IP) to all the BSs being adjacent to the one currently used. A combined multicast/Snoop scheme is reported in Ref. [27]. To reduce the increased traffic overhead associated with the solution reported in Ref. [27] and render our algorithm as efficient as possible, we suggest the stochastic relocation of datagrams according to the probabilities assigned to adjacent BSs by the path prediction algorithm.

In this paper we adopt the path prediction algorithm reported in Ref. [7] which is based on a learning automaton. However, the approach proposed here could use any other path prediction algorithm, provided that the latter assigns probabilities to the entire neighborhood of the present cell (i.e. does not simply identify the most probable cell but provides a vector of probability values). Additionally, to improve the efficiency of the algorithm, traffic duplication is only invoked after the MT spends a certain amount of time within the current cell. Such time is calculated through a low pass filter, by taking into account previous measurements of cell residence time (i.e. the time spent by the terminal in the current cell).

We have simulated the suggested architecture to evaluate the effects that stochastic datagram relocation has on TCP dynamics. We assume that BSs use a simple caching scheme, operating at the IP level, rather than the complex Snoop algorithm itself (which understands TCP dialogs). The simulation was performed using the MIL3 Opnet Modeler [15]. Through the simulation, we have tried to determine appropriate values for the basic operational parameters of the architecture (e.g. relocation percentages and invocation time). The obtained results, based on such parameter values, show considerable benefits for TCP connections.

The rest of the paper is structured as follows. In Section 2, we give an overview of prior work in the area of TCP enhancements for wireless/mobile communications. In Section 3, we discuss the details of the proposed scheme for wireless environments. In Section 4, we give a synopsis of the path prediction algorithm assumed for our study. In Section 5, we present the results of the simulation of the suggested relocation scheme. We conclude the paper in Section 6, where we outline directions for future work in the area.

Section snippets

Related prior work

Recently, a number of schemes have been proposed to cope with the transport protocol impairments over wireless links of high BER and many lost packets [17], [25]. Such schemes are classified into three main categories

  • the end-to-end approach,

  • the split-connection approach, and

  • the link-layer approach.

The first category involves, among others, some new TCP versions with enhancements on the standard protocol. This includes TCP Tahoe [20], which, upon reception of three duplicate acknowledgement

Architecture description

Fig. 1 shows the wireless network assumed in this study. The environment consists of a number of BSs interconnected by means of a fixed LAN infrastructure, as well as mobile and fixed terminals. Each BS comprises a radio transceiver and, possibly, a support workstation (the latter may handle all the signaling needed for the roaming of the MT). BSs maintain information regarding the MTs that are currently under their control. We also assume that each BS maintains a Datagram Relocation

Path prediction algorithm

The use of a PPA in a wireless network architecture allows the efficient use of limited network resources such as bandwidth. A number of PPAs have been proposed in the networking literature. Notable examples of such work are the algorithm proposed by Liu et al. [11] and the Liu–Maguire algorithm [12]. The former example uses pattern matching techniques and extended, self learning, Kalman filters to estimate the future location of MTs. The Liu–Maguire algorithm is based on Mobile Motion

Simulation results

A series of simulation experiments were realized in order to quantify the effects that stochastic relocation of datagrams has on TCP dynamics. We adopted a two-stage simulation approach. In the first stage, we tried to determine appropriate values for relocation percentages Y and Z for cells belonging to Classes B and C, respectively. Generally, we assume that a cell belonging to Class A receives 100% of inbound traffic (i.e. X=100%). After estimating the Y and Z percentages, we proceeded with

Conclusions and future work

In this paper we have proposed an architecture for overcoming the problems associated with the operation of TCP in wireless networking environments. We suggested the use of datagram/packet caches in BSs. Inbound traffic is stochastically relocated by the current BS to all its adjacent BSs and cached there. Our approach is aligned with the general design orientation of contemporary TCP enhancements: more functionality is introduced in the lower layers in order to meet the expectations of the

References (27)

  • R. Caceres et al.

    Improving the performance of reliable transport protocols in mobile computing environments

    IEEE JSAC

    (1995)
  • H. Balakrishnan et al.

    Improving TCP/IP performance over wireless networks

    Proceedings of the First ACM International Conference on Mobile Computing and Networking, (Mobicom)

    (1995)
  • E. Amir, H. Balakrishnan, S. Seshan, R. Katz, Efficient TCP over networks with wireless links, Proceedings of the Fifth...
  • A. Bakre et al.

    Handoff and system support for indirect TCP/IP

    Proceedings of Second Usenix Symposium on Mobile and Location-Independent Computing

    (1995)
  • K. Wang et al.

    Mobile-end transport protocol: an alternative to TCP/IP over wireless links

    Proceedings IEEE Infocom

    (1998)
  • S. Seshan et al.

    Handoffs in cellular wireless networks: the Daedalus implementation and experience

    Kluwer International Journal on Wireless Communication Systems

    (1996)
  • S. Hadjiefthymiades et al.

    ESW4: enhanced scheme for WWW computing in wireless communication environments

    ACM SIGCOMM CCR

    (1999)
  • M. Mathis, J. Mahdavi, S. Floyd, A. Romanow, TCP Selective Acknowledgements Options, RFC-2018,...
  • R. Yavatkar, N. Bhagawat, Improving End to End Performance of TCP over Mobile Networks, Proceedings of Mobile '94,...
  • D. Levine et al.

    A resource estimation and call admission algorithm for wireless multimedia networks using the shadow cluster concept

    IEEE/ACM Transactions on Networking

    (1997)
  • T. Liu et al.

    Mobility modeling, location tracking, and trajectory prediction in wireless ATM networks

    IEEE JSAC

    (1998)
  • G.Y. Liu et al.

    A predictive mobility management algorithm for wireless mobile computing and communications

    Proceedings of ICUPC '95, Tokyo, Japan

    (1995)
  • K. Narendra et al.

    Learning Automata: An Introduction

    (1989)
  • Cited by (0)

    View full text