ATP: A Reliable Transport Protocolfor Ad-hoc Networks∗Karthikeyan Sundaresan,†Vaidyanathan Anantharaman,§Hung-Yun Hsieh,†and Raghupathy Sivakumar††School of Electrical and Computer Engineering§Wipro TechnologiesGeorgia Institute of Technology India{sk,hyhsieh,siva}@ece.gatech.edu [email protected] works have approached the problem of reliable transportin ad-hoc networks by proposing mechanisms to improve TCP’sperformance over such networks. In this paper we show throughdetailed arguments and simulations that several of the design ele-ments in TCP are fundamentally inappropriate for the unique char-acteristics of ad-hoc networks. Given that ad-hoc networks are typ-ically stand-alone, we approach the problem of reliable transportfrom the perspective that it is justifiable to develop an entirely newtransport protocol that is not a variant of TCP. Toward this end,we present a new reliable transport layer protocol for ad-hoc net-works called ATP (ad-hoc transport protocol). We show throughns2 based simulations that ATP outperforms both default TCP andTCP-ELFN.Categories and Subject DescriptorsC.2.2 [Computer-Communication Networks]: Network Proto-colsGeneral TermsAlgorithms, Design, PerformanceKeywordsAd-hoc networks, Transport layer, Reliability, Rate adaptation1. INTRODUCTIONAd-hoc networks are uniquely characterized by the several fac-tors that differentiate them from traditional computer networks: (i)Lack of a fixed infrastructure: Due to absence of dedicated routers,mobile hosts in ad-hoc networks also serve as peer-to-peer relays∗This work was funded in part by NSF grants ANI-0117840 andECS-0225497, Yamacraw, and the Georgia Tech Broadband Insti-tute.Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.MobiHoc’03, June 1–3, 2003, Annapolis, Maryland, USA.Copyright 2003 ACM 1-58113-684-6/03/0006 ...$5.00.for connections in the network. (ii) Mobility: All hosts in the net-work are mobile, and hence the network topology can be highlydynamic. From the perspective of a single end-to-end connection,not only are the end-hosts mobile, but the intermediate “routers”are mobile too. (iii) Shared channel: Because of the all-wirelessnature of ad-hoc networks, not only do flows in the same vicinitycontend with each other, but part of a flow traversing multiple hopscan contend with other parts of the same flow in its vicinity. (iv)Limited bandwidth: While mobile hosts in general can be assumedto possess fewer amounts of resources than their static (wireline)counterparts, the wireless channel bandwidth is also scarce, result-ing in multi-hop flows typically enjoying limited bandwidths of atmost a few hundred kilobits per second.Not surprisingly, such constraining characteristics render tradi-tional wireline network protocols at the different layers of the pro-tocol stack inappropriate for use in ad-hoc networks. At the mediumaccess control (MAC) layer, protocols tailored for wireless environ-ments such as carrier sense multiple access with collision avoidance(CSMA/CA) easily outperform traditional protocols such as carriersense multiple access (CSMA). At the network layer, numerousrouting protocols such as dynamic source routing (DSR) [1], ad-hoc on-demand distance vector (AODV) routing [2], etc., have beenproposed for ad-hoc networks. Since such protocols are specificallytailored for the unique characteristics of ad-hoc networks, they sig-nificantly outperform conventional wireline routing protocols in anad-hoc network environment.At the transport layer, several works have focused on both study-ing the impact of using transmission control protocol (TCP) asthe transport layer protocol, and improving its performance eitherthrough lower layer mechanisms that hide the characteristics of ad-hoc networks from TCP, or through appropriate modifications tothe mechanisms used by TCP [3–10]. Given the almost universaluse of TCP as the transport layer protocol in the current Internet,such works are clearly warranted. However, several applicationsof ad-hoc networking, including more promising ones such as mil-itary battlefields, disaster relief operations, etc., are environmentswhere a completely revamped protocol stack tailored to the operat-ing conditions is not merely feasible, but also justifiable.In this paper, we approach the problem of providing reliabletransport over ad-hoc networks from the above perspective. Weargue that TCP or a minor variant of it is not appropriate for theoperating conditions common in ad-hoc networks. We study thesuitability of the different mechanisms used by TCP for congestioncontrol and reliability, for the characteristics of the target environ-ment. We discuss why a majority of the properties and mechanismsof TCP including window based congestion control, slow-start, lossbased congestion detection, multiplicative decrease of congestion64050010001500200025003000350040006262.5 63 63.5 64 64.5 65TCP Sequence NumberTime (sec)Sequence Number vs Time (1 connection)1 Flow Default TCP(a) [1 Flow, 1m/s]0200400600800100012002930 31 32 33 34 35 36TCP Sequence NumberTime (sec)Sequence Number vs Time (25 connections)1 Flow Default TCP(b) [25 Flows, 1m/s]Figure 1: TCP burstiness00.20.40.60.811.20 20 40 60 80 100RTT (sec)Time (sec)RTT vs Time (1 connection)1 Flow Default TCP(a) [1 Flow, 1m/s]00.511.522.50 20 40 60 80 100RTT (sec)Time (sec)RTT vs Time (1 connection)1 Flow Default TCP(b) [25 Flows, 1m/s]010203040506070800 5 10 15 20Max RTO (sec)Mobility (m/s)Average Max RTO: TCP-default1 Flow5 Flows25 Flows(c) RTO valuesFigure 2: TCP RTT samples and RTO valueswindow, use of retransmission timeouts, and reliance on reversepath characteristics, are not fundamentally appropriate for an ad-hoc network.We then present a new transport protocol called ATP (ad-hoctransport protocol) that is tailored toward the characteristics of ad-hoc networks. ATP, by design, is an antithesis of TCP and con-sists of: rate based transmissions, quick-start during connectioninitiation and route switching, network supported congestion detec-tion and control, no retransmission timeouts, decoupled congestioncontrol and