320 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 2, FEBRUARY 2004 PETRA: Performance Enhancing Transport Architecture for Satellite Communications Mario Marchese, Member, IEEE, Michele Rossi, Student Member, IEEE, and Giacomo Morabito, Member, IEEE Abstract—This paper presents a performance enhancing transport architecture for the satellite environment. This solution improves the network transport performance by overcoming the limits imposed by a transmission control protocol/Internet protocol (TCP/IP)-based stack suite, while maintaining the interfaces offered by it. This is an important issue since TCP/IP is widely used and most of the applications are based on it. The work starts from the state-of-the-art about the transport layer over satellite by distinguing two alternative frameworks: the black box (BB) and the complete knowledge (CK) approaches. In the former, the network is considered as a “black box” and only modifications in the terminal tools are permitted. In the latter, the complete control of any network element is allowed so as a performance optimization procedure is possible. The proposed ar- chitecture [called Performance Enhancing Transport Architecture (PETRA)] is designed in all details using the second approach. PETRA uses network elements, called relay entities, to isolate the satellite portions in case of heterogeneous networks, while a transport layer protocol stack is used to optimize the transport of information over satellite links. A special satellite transport protocol, that is part of the transport layer protocol stack, is used over such links to perform error recovery. Simulation results show that the proposed framework significantly enhances throughput performance. Index Terms—Delay effects, satellite communications, transport protocols. I. INTRODUCTION T HE transmission control protocol (TCP) is the connec- tion-oriented, end-to-end reliable transport protocol uti- lized in the Internet. The motivations, the philosophy, and the functional specification of the protocol are contained in [1]. TCP assumes that the service offered by lower layer protocols is unreliable; various features such as timeout timers, packet re- ordering, and retransmissions are used in TCP with the aim of providing a reliable channel to higher layer applications. Unfor- tunately, these features have been designed to be effective over wired networks, but it is well known that they often fail when the underlying channel is characterized by both a large bandwidth delay product and high error rates. This is the case of heteroge- Manuscript received December 15, 2002; revised July 1, 2003 and September 20, 2003. This work was supported in part by the European Space Agency (ESA) under Contract 14956/00/NL/ND. M. Marchese is with the CNIT-Italian National Consortium for Telecom- munications, DIST-University of Genoa Research Unit, University of Genoa, Genoa 16145, Italy (e-mail: mario.marchese@cnit.it). M. Rossi is with the Department of Engineering, University of Ferrara, Fer- rara 44100, Italy (e-mail: mrossi@ing.unife.it). G. Morabito is with the Dipartimento di Ingegneria Informatica e delle Telecomunicazioni, University of Catania, Catania 95125, Italy (e-mail: giacomo.morabito@diit.unict.it). Digital Object Identifier 10.1109/JSAC.2003.819981 neous networks containing satellite links, where special coun- termeasures have to be taken to correct the inefficiencies of the TCP protocol. Satellite networks are attractive since they offer clear advan- tages with respect to cable networks [2]: their architecture is more scalable, the diffusion throughout the land is wide, and the multicast service is very simple. Given the inefficiencies of stan- dard TCP over such networks, one could design a new protocol, specifically tailored for these environments. However, given the widespread diffusion of TCP/IP applications, it is very difficult to think of a protocol different from TCP, that is still transparent to the user, to be used over satellite links. For these reasons, it is more appealing to provide new solutions, where the TCP/IP can still be used at the end terminals and its inefficiencies are accounted by algorithms and protocols running at the end ter- minals and in the satellite portion of the network. The aim of this paper is to propose and validate a transport architecture to deal with this problem. In the sequel, a transport layer architecture that allows transporting TCP/IP flows efficiently and transparently through satellite networks to the final user is presented. The proposed architecture, which we call performance enhancing transport architecture (PETRA), divides the end-to-end connection into different segments. The bridging between different network segments is performed by elements named relay entities (REs). The objective of PETRA is the optimization of both the throughput performance for the satellite network environ- ment and the efficient utilization of network resources. This is achieved without redesigning the protocol interfaces, so that they maintain the same characteristics of the interfaces currently used. As a consequence, system performance is opti- mized and at the same time, utilization of standard applications is still possible, thus reaching a high degree of portability. Moreover, the proposed solution preserves the end-to-end reliability and semantic of the transport layer by introducing, at the transport layer, a new sublayer named upper transport layer. The transport layer is divided into two sublayers. • Lower Transport Layer (LTL): A different LTL instance is utilized in each segment of the end-to-end path. The LTL is responsible for the transport in each of these segments. The LTL utilized in the satellite segment is called satellite transport protocol plus (STPP). It is similar to the STP proposed in [3] with the following differences. — STPP introduces a mechanism that avoids deadlock sit- uation. — STPP introduces a mechanism which stops the flow over the satellite link to avoid buffer overflows in the REs. 0733-8716/04$20.00 © 2004 IEEE