1

ACP-WG I-02/WP-09
/
International Civil Aviation Organization
WORKING PAPER / ACP-WG I-02/WP-09
23/08/07

AERONAUTICAL COMMUNICATIONS PANEL (ACP)

SECOND MEETING OF THE WORKING GROUP I (IPS)

Montreal, Canada 27 – 31 August 2007

Agenda Item xx: / Xxx

TRANSPORT LAYER CONSIDERATIONS IN THE ATN/IPS

(Presented by Frank Schreckenbach, Daniel Medina, Christian Kissling)

SUMMARY
This paper discusses transport layer issues and their implication on network layer technologies for the air/groundATN/IPS, as investigated in the NEWSKY project. The performance degradation of connection-oriented transport layer protocols such as TCP in wireless heterogeneous environments motivates the search for an efficient end-to-end protocol architecture.
ACTION
To review the material in this paper and develop further guidance and action, as required.

1.INTRODUCTION

In the current ATN/OSI, two transport layer protocols have been specified for use in ATN End Systems[1]: the Connection Oriented Transport Protocol (COTP) and the Connectionless Transport Protocol (CLTP). A given End System (ES) may implement one or both of these, depending upon the requirements of the applications it contains.

COTPprovides to its users an end-to-end connection mode service, and is a conformant subset of the Class 4 Transport Protocol (TP4) specified in ISO/IEC 8073. The tasks of the COTP protocol include the provision of a reliable connection over unreliable lower layers and the provision of flow control and error control in case datagrams are lost. The COTP protocol splits the data streams of different applications, assigns a header to them and hands the data over to the network layer. When receiving network layer datagrams, the COTP protocol orders the received datagrams and reconstructs the original data stream. If packets are lost, the COTP protocol recognizes this and requests these datagrams again. This ensures both data integrity and data sequence integrity.

CLTP provides a connectionless transport service, where no service guarantees are offered. Each message is transferred as an event unrelated to the transfer of any other message and there is no guarantee either of delivery or that a given message may not overtake an earlier message. The CLTP is conformant with ISO/IEC 8602.

Similarly, in the IP world, the most popular transport layer protocols are TCP (Transmission Control Protocol)[2] and UDP (User Datagram Protocol) [3]. TCP guarantees reliable and in-order delivery of data from sender to receiver, while with UDP, datagrams may arrive out of order or go missing without notice.

While transport layer protocol use in the ATN/IPS is still an open issue, the expectation is that both TCP and UDP will find their role in the ATN/IPS, depending, as in the current ATN/OSI, on each application’s requirements.

Although regular versions of TCP work fine in the fixed Internet, their performance degrades significantly in the wireless Internet, since the assumptions of short delays and very low BER that were made during its design for the former are no longer valid in the latter.

The properties of wireless networks, especially the satellite link differ significantly from these assumptions. Some properties of satellite links are:

  • long and very long delays
  • delays are variable over time, especially for non-geostationary satellites
  • large delaybandwidth product
  • higher BER, especially for mobile terminals
  • losses and fading
  • high asymmetry of data rates in the uplink and downlink

Further considering the example of a satellite link, the following list provides a non-comprehensive overview of problems arising when using a regular version of the TCP protocol for this link:

  • Connection establishment with handshaking takes very long.
  • The TCP slow-start phase takes very long. This means that for short-duration connections the link is used in a highly inefficient manner. This especially affects application sessions which use several short TCP connections throughout a single application session.
  • Flow Control: Due to the credit based Flow Control mechanisms in TCP the throughput is limited depending on the round trip delay.
  • Congestion Avoidance: In the terrestrial internet segment losses are usually caused by an overload of the network. For this reason TCP reduces the congestion window for each lost segment to reduce the network load (congestion control). However, the TCP protocol does not distinguish whether segment losses occur due to congestion or link loss phenomena. In the satellite link, bit errors also cause the loss of segments and lead to a transmitter time-out. The following slow-start and congestion avoidance phases reduce the throughput on the link significantly. In addition, the retransmission timer is increased (RTO back off). Due to the long propagation delays of the satellite link many data segments have to be retransmitted and it takes long until slow-start and congestion avoidance phases are completed.

The ATN/IPS is by definition a wireless internet, and any potential transport layer protocol should take this into account. In addition, in the ATN/IPS, mobile nodes (aircraft) can communicate with correspondent nodes on the ground (e.g. ATCCenters, AirlineOperationsCenter, Public Internet…) via multiple wireless links (depending on their location and equipage) such as Gatelink, WiMAX, VDL Mode 2, B-AMC, P34, Inmarsat SwiftBroadband, DVB-S2/RCS, to name a few candidates. This scenario is shown in a simplified fashion in Fig. 1. This makes the problem even more complicated, since each potential wireless access technology hasdifferent characteristics, affecting the performance of TCP in a different way.

To summarize, we are confronted with the problem of enhancing TCP performance in a heterogeneous wireless environment, where each link has its particular idiosyncrasies.

These transport layer issues have to be tackled at the very start of the network development phase since they strongly influence most network layer issues, including, mobility, routing, and security.

Fig. 1 ATN/IPS End-to-End Architecture

2.ENHANCING TCP PERFORMANCE IN A HETEROGENEOUS WIRELESS ENVIRONMENT

This is a complex task, and may be decomposed intuitively into two subtasks:

  1. TCP needs information about the characteristics of the link used to transmit the packets, so that it can personalize its behaviour for that specific link.
  2. Once it is aware of the specific characteristics of the link in use, how can TCP personalize its behaviour to function optimally over that specific link? In other words, how can TCP performance be enhanced over that link?

2.1Keeping TCP informed of link layer events

Such a heterogeneous wireless environment is not only found in aviation, but also in the context of modern mobile devices, such as those considered in the IEEE 802.21 standard (“Media Independent Handover”)[4]. These devices are equipped with a variety of access technologies, such as 802.11 (WLAN), Bluetooth, 3GPP, 3GPP2, WiMAX, etc. The scope of IEEE 802.21 is to develop a standard that provides link layer intelligence and other related network information to upper layers (e.g. IP, Mobile IP, SIP, TCP, UDP, application layer) to optimize handovers between heterogeneous media.

To illustrate how this standard relates to the problem at hand, consider these examples:

  1. The Link_Up.indication primitive notifies upper layers when a layer 2 connection is established for the specified link interface. When higher layer connections are being established over this link, the transport layer may adjust transport (TCP) related parameters based on link properties.
  1. The MIH_Link_Handover_Complete.indication primitive notifies upper layers when a link layer inter- or intra-technology handover is completed. Transport layers (e.g., TCP) may make use of this primitive to fine tune their flow control and flow congestion mechanisms.

A solution analogous to IEEE 802.21 is expected to be realized in aviation to optimize handovers between heterogeneous media such as satellite and terrestrial radio links. TCP parameter adaptation could benefit from the presence of this functionality.

However, the functionality provided by 802.21 only solves subtask a above. It keeps TCP (among others) well informed about what is going on at the link level. A solution is also needed for subtask b, indeed for each particular link in turn.

2.2Enhancing TCP performance over wireless links

Several approaches have been proposed to enhance the performance of TCP over wireless links

Split connection approaches like I-TCP [6] and MTCP [7] propose to split the TCP connection between the fixed host (FH) and mobile host (MH) at the intermediate base station (BS).The split connection approaches propose to have a protocol optimized for wireless links to enhance performance of TCP. This mechanism is typically implemented by a so-called Performance Enhancing Proxy (PEP) located in the base station and mobile host.

RFC 3135 [5] is a survey of existing PEP solutions that include different approaches for improving TCP performance and their implications.

Other approaches to improve performance of TCP like SNOOP [8], M-TCP [9], WTCP [10], ELN [11] preserve end-to-end semantics of TCP. However, all these approaches require base station mediation and are based on the assumption that TCP headers are readable by the base station.

2.2.1Coexistence of IPsec end-to-end security and TCP performance enhancements

The above mentioned approaches supplement TCP either by explicitly providinga mechanism for detecting the nature of lossor by shielding the wireless losses from the TCP sender.However, the mechanisms are tightly coupled withthe transport layer, which inhibit the use of securitymechanisms at network layer.Achieving improved TCP performance together with ensuring end-to-end security necessitates the co-existence of security mechanisms like IPsec and performance enhancing solutions. However, IP security and TCP performance have been traditionally dealt with in a mutually exclusive manner.

IPsec providessecurity services at the IP layer and can beused to protect one or more paths between a pairof hosts, a pair of security gateways or between asecurity gateway and a host [12]. The security at theIP layer provides protection to both the applicationdata or payload and the transport layer headers (TCP or UDP headers). IPsec uses two protocolsto provide the security services, namely AuthenticationHeader (AH) [13] and Encapsulating SecurityPayload (ESP) [14]. AH may be used to provide dataintegrity, origin authentication and anti-replay protection.ESP provides the entire authentication servicesprovided by AH and along with it, it alsoprovides confidentiality or encryption.

In the ATN/IPS, IPsec plays a vital role, guaranteeing end-to-end security between aircraft and their correspondent nodes (as well as between different organizations within the ground internet). Therefore, a solution must be found to enhance TCP performance over the various access technologies without the respective base station or satellite gateway requiring access to the TCP headers of packets exchanged with the aircraft, since these will be encrypted by IPsec end-to-end. This is currently a hot topic in the Internet research community [15].

3.REFERENCES

[1]“Comprehensive ATN Manual, Part IV Communication Services”, January 1999.

[2]IETF RFC793 “Transmission Control Protocol (TCP)”, September 1981.

[3]IETF RFC768 “User Datagram Protocol (UDP)”, August 1980.

[4]IEEE P802.21/D05.00 “Draft Standard for Local and Metropolitan Area Networks: Media Independent Handover Services”, April 2007.

[5]IETF RFC3135 “Performance Enhancing Proxies Intended to Mitigate Link-Related Degradations”, June 2001.

[6]A. Bakre, B.R. Badrinath, “I-TCP: Indirect TCP for Mobile Hosts”, in Proceedings of 15th International Conference on Distributed Computing Systems, May 1995.

[7]R. Yavatkar, N. Bhagawat, “Improving end-to-end Performance of TCP over Mobile Internetworks”, in Proceedings of IEEE Workshop on Mobile Computing Systems and Applications, December 1994.

[8]H. Balakrishnan, S. Seshan, E. Amir, R.H. Katz, “Improving TCP/IP performance over wireless networks”, in Proceedings of ACM Mobicom, November 1995.

[9]K. Brown, S. Singh, “M-TCP: TCP for mobile cellular networks”, ACM Computer Communications Review (CCR) 27 (1997) 5.

[10]K. Ratnam, I. Matra, “WTCP: an efficient mechanism for improving TCP performance over wireless links”, in IEEE Symposium on Computers and Communications, June 1998.

[11]H. Balakrishnan, R.H. Katz, “Explicit loss notification and wireless Web performance”, in Proceedings of IEEE Globecom Internet Mini-Conference, November 1998.

[12]IETF RFC4301 “Security Architecture for the Internet Protocol”, December 2005

[13]IETF RFC4302 “IP Authentication Header (AH)”, December 2005

[14]IETF RFC4303 “IP Encapsulating Security Payload (ESP)”, December 2005

[15]V. Obanaik, L. Jacob, A.L. Ananda, “Secure performance enhancing proxy: To ensure end-to-end security and enhance TCP performance over IPv6 wireless networks”, September 2005.