Behavior of TCP-Probing with Hand-offs.
Adrian Lahanas, Vassilis Tsaoussidis · International Conference on Internet Computing · 2001
Abstract: Performance of standard TCP is signifi-cantly degraded during hand-off periods of mobile commu-nications. We show here that carefully designed probingmechanisms can cancel this incompetent behavior of TCPfor a wide range of hand-off periods. We compare TCPTahoe and Reno with TCP Probing under various scenariosof hand-off duration, propagation delay and mobility con-ditions of the user. We measure the goodput efficiency ofdata transfer as well as the energy efficiency of the proto-cols’ mechanisms in terms of overhead and time. Our re-sults show that TCP Probing demonstrates significant gainsin both energy and throughput efficiency by avoidingunnec-essary timeout extensions and retransmission of data whichis likely to be lost due to hand-offs. 1 Introduction TCP is fine-tuned for wired networks where link er-rors and mobile-specific operations is not a concern.Missing packets or lack of acknowledgments by thereceiver is interpreted as congestion and TCP appliescongestion control algorithms [1]. In mobile comput-ing where networks consist of wireless cells and usersmightmove betweenthe cells, TCPexperiencesdelaysdue to hand-offs. The connection goes idle as long asthe mobile host and the Base Station are involved inMobile IP Routing [2]. This delay is interpreted ascongestion by TCP and as a consequence its perfor-mance is degraded significantly.The performance of TCP in mobile computing en-vironments has been studied thoroughly by several re-searchers. Caceres and Iftode [3] have studied the im-pact of hand-offs on TCP and in order to improve itsperformance they proposed a three DACK reply fromthe mobile host to the TCP sender immediately afterthe hand-off termination. Zorzi and Rao [4] have stud-ied the performance of TCP over wireless links withfading channels and found that most of the TCP imple-mentations (Tahoe, Reno and New-Reno)have similarenergy and throughputperformancewiththe exceptionof high error rates where Tahoe yields better results. Ingeneral he observed that time-outsare extended duringthe hand-off process or when the link is fading.Solutions that deal with the impact of hand-offson TCP performance can be grouped in three cate-gories: (i) split TCP solutions that correspond to theparticularity of the network (i.e., wired/wireless), (ii)proxy solutions that require intervention at the basestation and (iii) end-to-endsolutions at the transportlevel only. A proposal of a split protocol is Indirect-TCP [5]. An example of a reliable protocol in thesecond category is the Snoop-TCP[6]. The proto-cols in the two first categories need intervention at thebase station and once they are deployed it is practi-cally difficult to change or upgrade them. In addi-tion, these solutions do not comply with the universalnature of an end-to-endprotocol such as TCP whichis designed to work with a variety of heterogeneousnetworks and devices. Protocols in the third categorysolve the problem by modifying the transport proto-col at the end-hosts,maintaining so the end-to-endse-mantics of TCP. A novel approach in this category isthe Freeze-TCP[7] where the transport layer can in-teract with all the layers below the stack. As soon asthe receiver receives a signal for a potential hand-offprocess, TCP receiver (i.e., here the mobile host) ad-vertisesa zero window tothe sender. This approach re-quires that the hardware detects in advance a potentialhand-off and notifies the transport layer. Another ap-proach in the same category is TCP-Probing [8]. Thecore mechanism of this protocol is the Probing Cyclewhich is initiated every time the sender experiences1