Performance Evaluation of an Ultra-Fast Pipeline Scheduler for Next-Generation Networks M. T. Anan*, G. M. Chaudhry* and J. Qaddour ** *School of Computing and Engineering, Univ. of Missouri-Kansas City ** School of Information Technology, Illinois State University mtanan@umkc.edu, chaudhryg@umkc.edu, & jqaddou@ilstu.edu Abstract  The Internet Protocol (IP) plays a dominant role in current networking technologies and the Wavelength Division Multiplexing (WDM) technology provides tremendous bandwidth. The IP-over-WDM is becoming the right choice for Next-Generation Internet networks where Optical Burst Switches (OBS) is a promising technique to bridge the gap between IP and WDM. However, traditional OBS scheduling algorithms either have low computational complexity with high burst dropping probability or high computational complexity with low burst dropping probability. A critical design issue in OBS is how to reduce burst dropping probability as a result of resource contention using a high performance scheduling algorithm. In this paper, we present an ultra-fast scheduler which integrates the merits of both low computational complexity and low burst dropping probability. The key idea is to maintain all scheduled data bursts and void intervals in a binary representation. Then, fast hardware logic operations are performed to schedule incoming data bursts without the need to traverse and search all scheduled bursts in all channels. The new approach provides an effective optical burst switching for variable-length bursts with low computational complexity compared to existing scheduling algorithms. Index Terms – Optical Burst Switching, Core Node Architecture, Wavelength Conversion, Channel Scheduling, Contention Resolution, Optical Fiber Delay Lines I. INTRODUCTION Optical burst switching (OBS) is a promising technology to address the problem of efficiently allocating resources for next-generation optical switched networks [1]. However, this technology relies on statistical multiplexing in order to achieve good utilization in presence of bursty traffic. As a consequence, contention situations occur due to a reservation or transmission conflict. This leads to high burst loss. There has been a large amount of research work addressing the issue of OBS resource contention in recent years [2-4]. Architectures with complicated scheduling algorithms are less appealing especially in high-speed optical networks. In OBS, an ideal scheduling algorithm should be able to process a control packet fast enough before its corresponding data burst arrives, and yet be able to find a suitable time interval to schedule the burst. Otherwise, a burst may be discarded either because a reservation cannot be completed before the burst arrival or simply because the scheduling algorithm is not smart enough to make the appropriate reservation. Hence, a key challenge is to design an efficient scheduling algorithm for bandwidth reservation to minimize burst loss probability. This paper introduces an ultra-fast scheduling scheme to minimize the burst loss probability and scheduling delay. This paper is organized as follows: Section II describes the basic components and operation of an OBS switch. Section III briefly reviews some of the most commonly used scheduling algorithms for OBS. Section IV describes our proposed approach. Section V discusses the performance of the proposed approach and compares it to the most famous scheduling algorithms for OBS. Finally, section V concludes the paper. II. BACKGROUND In OBS networks, control signaling is performed out- of-band by having few channels to be dedicated to carry headers or Burst Control Packets (BCP). Only these channels go through Optical/Electronic/Optical (O/E/O) conversion. Data transparency is achieved by switching data bursts all-optically at burst level. This helps achieving good utilization of network resources in presence of bursty traffic. Several OBS core node architectures have been discussed in the literature [5-8]. These architectures vary in design and hardware complexity. In general, OBS core nodes have the general architecture which consists of two parts: the switch control unit (SCU) and the data burst unit [9]. The SCU is responsible for processing and interpretation of BCPs, scheduling, collision detection and resolution, forwarding table lookup, switching matrix control, header rewrite, and wavelength conversion control. The data burst unit is responsible for switching data bursts to the destined output port in the optical domain. The core node could be equipped with fiber delay lines (FDLs) to store data bursts optically. The storage time varies depending on the length of the FDLs. Fig. 1 shows the architecture of a generic 2x2 OBS core node architecture together with the contention resolution components. The switch architecture has two input and two output fibers each fiber could have n wavelengths for data channels and c wavelengths for control channels. The switch is composed of both optical and electronic components. 650 1-4244-1031-2/07/$25.00©2007 IEEE