A Real-time TCP Management Scheme over OLS Networks Yassine Khlifi and Noureddine Boudriga CN&S Research Lab., University of November 7th at Carthage, Tunisia Corresponding author: nab@supcom.rnu.tn ABSTRACT To support the fast growing demand for transmission capacity, optical label switching technology (OLS) seems to be a strong candidate due to its ability to allow fast switching and resources provisioning as well as quality of service (QoS) support. Since TCP is and may remain the most popular traffic in Internet, TCP implementation becomes a major concern in OLS networking and operation. To take full advantage of OLS networks and integrate several TCP functions in the optical layer, the implementation of TCP over novel OLS network is addressed for providing QoS control, resources provisioning and finer granularity analysis. In this paper, we introduce a novel technique for packet retransmission based on real-time management of TCP parameters, network resources and traffic requirements. The technique handles activities including signaling, switching tasks and contention resolution. We develop a scalable packet retransmission approach based on buffering technique and contention resolution, and formalize it using network resources availability in terms of buffering and transmission capabilities available on the established optical path. Finally, simulation activity is conducted to validate the proposed approach and evaluate its performances. I. INTRODUCTION The fast growing of Internet traffic and the emergence of new applications has introduced new requirements in terms of resources utilization and quality of service (QoS) support. Optical label switching (OLS) has been proposed as a new paradigm to fully utilize the benefits of dense wavelength division multiplexing (DWDM) and provide fast switching capability and QoS support as well as resource provisioning [1, 3]. In such a network, the traffic is transmitted using a set of labels created during an out- of-band signaling scheme, which uses a two-way signaling scheme to build the label switching path (LSP) and update the switching tables of LSP-cores. QoS control and resource provision are critical tasks in OLS networks, since resources management and efficient data transmission costs are essential design criteria in networking platform. While TCP is the dominant transport protocol today and likely to be adopted in future optical networks, TCP implementation becomes an attractive topic that can be addressed for enabling several fundamental functions in optical layer such as QoS control, finer granularity analysis and efficient data transmission costs as well as resources provisioning. Currently, TCP implementation over optical networking systems is an important design issue, especially in OLS networks. To perform advanced management levels of available resources and QoS control, we consider in this work an OLS network where all nodes perform a dynamic service differentiation using traffic requirements [3]. To our knowledge, there is no previous work related to TCP implementation over OLS networks integrating buffering process and dynamic contention resolution. In most of existing works, authors have attempted to implement some of TCP functions over optical burst switching (OBS) networks using just-enough-time (JET) signaling scheme to resolve false time-out detection (FTO) and false congestion detection (FCD) problems. These methods provided a static approach of time-out and congestion detection parameters [4]. Other works (e.g. [5, 6]) have addressed burst delay and end-to-end TCP throughput using a core node architecture with no optical delay lines (ODL) and a limited wavelength conversion capability. However, all these methods do not take into account ODL buffering due to the contention and real- time variation of system parameters, and cannot achieve a suitable decision for providing high level resources provisioning and QoS control. In this work, we address the issue of TCP implementation over OLS networks based on a dynamic real-time management of TCP parameters, and network resources such as buffering and transmission capabilities. We propose a scalable packet retransmission approach and formalize it using the available network resources in terms of transmission and buffering capabilities. We first extend the existing OLS signaling protocol in order to introduce two concepts: network resources management and dynamic TCP parameters control. The proposed approach considers a dynamic real-time supervision of core node resources and transmission link capability, as well as traffic characteristics, such as the real-time variation of time-out of the incoming traffic and buffering capacity at each core node related to the pre-established lightpath. We also develop an analytic model for formulating the proposed approach through the use of a conservation law and queuing network model [3]. Finally, simulation experiments have been conducted in order to validate the efficiency of our proposed approach. The remaining part of this paper is organized as follows. Section 2 briefly studies the basic concepts of the core node and network architectures. It also presents the adopted signaling and contention resolution schemes. Section 3 presents in detail the proposed packet retransmission protocol and its algorithm. Section 4 develops a theoretical model that helps the design and formulation of the proposed approach using a conservation law [9] and queuing network model. Section 5 presents the developed simulation model. It also