Hamdy Soliman Title: Associate Professor Affiliation: Computer Science Department, New Mexico tech Address: CS Dept. New Mexico tech Socorro, NM 87801 Email: hss@charon.cs.nmt.edu Tel: 505/835-5170 Fax: 505/835-5587 All Fiber Optics Wide Area Network With Lightware Routing Dedicated Global Wavelengths By Hamdy S. Soliman INTRODUCTION: A very significant contribution to the "Next Generation Internet" initiative is to move the existing internet routing delay from the software/hardware (protocols/ electronic) to the lightware (optical switching) delay, which will increase the Internet speed in the order of 1000 times. This can be done through the adaptation of "all fiber optics", AFO, internet backbone, with data moving at the speed of light through fiber switching only. The AFO networks approach has been investigated extensively in the literature, and few lab prototypes single-hop Wave Division Multiplexing (WDM) LANs have been introduced (e.g., IBM Rainbow-I&II, Bell Communication LAMBDANET, FOX). The main goal is to avoid any intermediate packet inspection (soft/ hardware) which forces a costly Optical-to-Electrical, OE, and vice versa, EO, conversions. Instead, routing will be done over preallocated fiber wavelengths, WL, end to end over fixed light paths with nano second delay. Taking advantage of the abundant fiber capacity of about 25 THz and low-attenuation in single-mode passband, the fiber bandwidth ,BW, is divided into about 150 separated WLs (150 only because of current technology restrictions). Data can be multiplexed over the same fiber line from one or more user(s) at different WLs, channels, each has capacity of few Gbps, allowing for extremely large total capacity networks (and user links), in a WDM architecture. Based on results from current researches and the recent development in the fiber domain (both, connections speed and cost), we are investigating the development WDM Wide Area Network (WAN) to work as the next generation "speed of light" internet backbone. Our new WAN consists of a group of broadcast&select passive "single hop" stars touching arms through fiber switches to allow up to 64 Local Area Networks (LANs), each of up to 32 nodes (bridge/user) to be interconnected with no EO/OE delay (only at the end stations) The system is still a prototype to be expanded, later, to a real cost-to-cost WAN amenable for "real time" applications over the future NGI project. Our design faces many challenges, such as, signal noise/attenuation over the WAN huge distance, current technology limitations and cost of the fiber optics components, but time is in our side. Also, the star reliability and robustness. A solution to the last challenge is to combine the star and loop topologies in a new pizza topology. If the star coupler goes down, the loop topology can be used (still in the fiber domain) until we repair the star. OUR MODEL: The core of our design is a 1XN passive star (which can get up to about 1X64 coupler and the use of the super high power single mode lasers). Among the many good reasons for selecting a single hop passive star technology, as the CPS versus other peer popular topologies , e.g., the linear bus, leaf (bus of stars), and ring topologies, we list two: 1) its large fan-out (number of arms), which allows for connecting N LANs using only N-1 WLs. 2) its smaller power loss (as a function of number of connected nodes). 3) its passive unpowered core 1XN muxing/dmuxing Our new WAN model consists of a central passive star, CPS, connecting other remote passive stars (RPS), LANs, via its arms, in a two-level structure, (we are also looking at a recursive multi-level structure, where any RPS can be viewed as CPS in a symmetric topology). The CPS total BW is divided into n regions, each with n-1 WLs, where n is the number of connected RPS (i.e., star LANs). Notice that n(n-1) can not exceed the total useful (possibly utilized) fiber WLs. Each of the CPS arms, which is connected to one of the n RPSs, will carry the total global traffic at each RPS, to all other RPSs, over one of the uniquely assigned regions (explained above). This separation of WLs at the CPS allows for fast lightware switching through the CPS core, with no interference and minimal routing protocols for the WAN global traffic. The total RPS star BW is divided over different types of traffic as follows: L, G, and C distinct WLs (channels) for the local, global, and traffics control, respectively. We will focus mainly on the global G channels assignment, the local and control channel assignment is well covered in the literature and some (or all) work can be utilized in our WAN. The type of global WLs assignment at each of the RPSs depends on what type of receivers to be used, fixed (cheap) versus tunable (expensive), our choice is n-1 fixed receivers to extract global traffic which is directed to the RPS LAN from the other n-1 peers. THE CPS BANDWIDTH DIVISION AND RPS GLOBAL WAVELENGTHS ASSIGNMENT: The division of the total useful CPS BW WLs is a simple process. For example, if we have 1X4 CPS, each of the connected RPS needs 3 WLs (window) into the CPS core BW. Thus, a total of 12 different WLs are needed to carry the total WAN global traffic over the CPS (with no intersection). the WL assignment is to be done linearly, assigning the first RPS LAN (attached to the CPS first arm) WLs 1, 2, and 3. The second RPS is to be assigned WLs 4, 5, and 6, and so on, until the fourth RPS which is to be assigned the WLs in the region 10, 11, and 12. Hence, each of the RPS nodes needs to be equipped with three fixed transmitters for the three assigned global WLs. Now, to the story of receiving global traffic from other RPS through the core of the CPS, over the WLs (one/LAN). If we build a table 4X4 index with the LAN numbers: L1 L2 L3 L4 L1 - 1 2 3 each table entry (i,j) is the assigned global L2 4 - 5 6 WL between LANi ---> LANj L3 7 8 - 9 L4 10 11 12 - it would be easy to figure out for any RPS node, say LANi, what are the WLs to be filtered (extracted out of the total light coming from the CPS). Simply, it is the column under that RPS LAN number (i.e., i for LANi). For example, LAN3 needs to mux WLs 2, 5, and 12 into its LAN, say over one dedicated WL, mainly assigned for global traffic reception, where all RPS nodes are having fixed receiver over it (a node will extract any traffic with its address as destination). The above example can be generalized for any 1Xn CPS connecting N LANs. The model can be expanded, recursively, into a star of stars, where each of star arm is to be connected to another star, of the same quality nodes.