Operation Joint in Multi-Hop WDM Networks under Dynamic Data Traffic*
Wu Kai, Zeng Qingji, Xiong Yizhi
R&D Center for Broadband Optical Networking Technology Shanghai Jiao Tong University, Shanghai 20030, P. R. China Email: [e-mail] Tel: [phone] Fax: [phone] Abstract Confronted with dynamic data traffic nowadays, we hereof raise and analyze the critical issues in joint operation under an IP over WDM architecture. This paper highlights the grooming issue, further formulates it into ILP, and finally proposes heuristic algorithms. Key Words multi-hop WDM network, traffic grooming, virtual topology, joint operation IP Adaptation 1. Introduction WDM (Wavelength Division Multiplexing) technology has been introduced into todays network infrastructure to accommodate the rapid growth of Internet data traffic. A new network layer incorporating OXCs (optical crossconnects) and OADMs (optical add/drop multiplexe rs) is evolving at the bottom of b ackbones in WAN and metropolitan networks. Recent researches are focused on its design and operation to get it prepared for future IP centric data traffic. Several architectures of IP over optics are then proposed [1-2]. To minimize total network cost and maximize network resource efficiency, two main approaches are proposed: (1) reducing the protocol stack; and (2) bypassing or eliminating electronic processing. As indicated in [1], network capacity is always limited by the electronic bottleneck. The reduction of intermediate protocol layers (e.g. ATM, SDH/SONET) greatly increases bandwidth efficiency, simplifies control and management, relieves electronic The focus of this paper is on the operation of this interlayer manager, namely joint operation of IP and WDM layers. In Section 2, we discuss in detail the network model and grooming issue. In Section 3, we formulate the grooming issue as an Integer Linear Program (ILP). Section 4 presents some heuristic algorithms and the numerical results, while Section 5 draws conclusions. 2. Network Model and Grooming Issue The WDM layer provides upper layer with a set of end-to-end transparent wavelength paths. Seen by the IP layer, these lightpaths constitute a virtual topology, which differs from the physical topology, consisting of optical nodes interconnected by fibers. By altering these lightpath connectivity between
*
processing load and hence reduces cost. A simplified protocol stack is proposed herein, as illustrated in Figure 1. restoration mechanism. The IP adaptation layer It also integrates an defines the framing, bandwidth provisioning and interlayer manager, which serves as a coordinator. IP Layer
Data Traffic
Interlayer Manager
Layer
Path Figure 1. IP over WDM protocol stack and the operation scheme.
WDM Layer
Wavelength
This work is jointly sponsored by National 863 Plan, National Natural Science Foundation and National Key Lab of Broadband Optical Fiber Transmission and Communication system
electronic switches, the WDM network provides the ability to change the virtual topology in response to traffic condition changes. The authors in [3] offer a comprehensive survey in virtual topology design problem. The problem is proved to be NP hard and typically decomposed into several sub-problems: (1) determination of virtual topology; (2) lightpath routing; (3) wavelength assignment; and (4) traffic routing. The grooming issues (sub-problem (1) and (4)) are tasks of interlayer manager depicted in Figure 1; while the RWA problems (sub-problem (2) and (3)) are addressed in the WDM layer, leaving some constraints to the former ones. The work in [6] discusses maximum wavelength reusability limited by average nodal degree of a given physical topology, which acts as an example of the above-mentioned constraints.
forwarding. Current work on grooming issue [4] -[5] is based on SDH/SONET over WDM architecture, and mostly rings. In this model, SONET ADMs dominate total network cost. In [4] a set of heuristic algorithms are provided to minimize both SONET ADM number and wavelength number. Non-uniform traffic is also considered in [4]. The authors in [5] enumerate various SONET ring architectures, and obtain some theoretical bounds. Only uniform traffic is considered in [5]. However, when the electronic equipments are IP switches/routers, some assumptions do not hold any more. The physical topology is not necessarily ring, but may be arbitrary. The objective of network optimization is not to minimize ADM number but to minimize electronic forwarding, so as to relieve electronic processing burdens. More flexible flow rates are assumed. As shown in Figure 2b, flows of various rates to different destinations (dynamic traffic) are encapsulated into certain lightpath. IP Layer Grooming/Routing
Traffic flows from application layer
IP Switch/Router Fra ming/Grooming Aggregate data at lightpath speed Add/drop Ports WDM OXC Figure 2a Node architecture. RWA algorithm O/E conversion Grooming Lightpath algorithm
E-layer resources
Virtual Topology = lightpath set O-layer resources
Lightpath Setup/Release Flow i , rate ) (i Figure 2b Aggregate data in one light -path. Grooming is efficiently multiplexing lower speed traffic into higher speed lightpaths to maximize optical bypass and hence minimize electronic WDM Layer Figure 3. Joint operation of IP and WDM layer
Figure 2a depicts the node architecture under protocol stack mentioned in section 1. It is composed of an IP switch and a WDM OXC. Generally, an OXC can
support several IP switches by assigning add/drop ports. A flow may follow a route in the sense of IP network, i.e. electronic forwarded by several IP switches at different nodes. A route goes through a concatenation of lightpaths (multi-hop). The joint operation of the network is shown in Fig3. The interlayer manager, highlighted in the shadow area, maintains a virtual topology graph. The optical layer resources include working & spare wavelengths & fibers. electronic The layer resources include occupied and spare bandwidths. These resources respectively decide the routes of lightpaths and data traffic flows, given the corresponding routing algorithms within each layer. With interlayer manager, the upper IP layer exchanges such traffic information as variation of flows (addition, removal, rate change etc.), while the lower WDM layer exchanges the configuration of lightpaths (setup, release, configurability, etc.). By implementing some reasonable grooming and RWA algorithm, the interlay manager ensures efficient use of both e-layer and o-layer resources. To note, Figure 3 only depicts the conceptual model of interlayer manager, which does not cover the implementation. The management entity or protocol executor could be integrated in either central or distributed network management systems. 3. Formulation of Grooming In formulation of the grooming issue, the following notations have been used: G(N,P) The graph denoting the virtual topology, where N is the set of network nodes, P is the set of lightpaths {p} for each s -d pair. R Set of routes {r}. A route r goes through a concatenation of lightpaths. F Set of flows { f}. A flow would choose a route to carry on traffic. To simplify the problem, we assume, only one flow chooses each route. Then we have, Variable, which denotes the rate of flow on r route r. (r[0,1], we assume each lightpath rate is 1). S Set of s-d pairs { s}. For simplification, we suppose all traffic flows and lightpaths are bi-directional ts Total traffic for s-d pair s. T is the set of ts.
Tp
Total optical capacity for lightpath p. Since a lightpath could be assigned with different wavelength, this value is an integer (may be greater than 1). A(r,s) Boolean matrix, taking on the value of 1 if route r belongs to s-d pa ir s, 0 otherwise. B(r,p) Boolean matrix, taking on the value of 1 if route r goes through lightpath p, 0 otherwise. Suppose S, P, F, T are given. The traffic routing based on current P is computed by some routing protocol or algorithm, i.e. A(r,s) and B(r,p) are also computed. Then we obtain the constraints from e-layer and o-layer respectively as,
A(r , s)
rR r R
r
= ts , Tp,
s p
(1) (2)
B (r, p)
Minimize
r
The objective is to minimize the electronic forwarding
T forward = r (h r 1)
r R
(3)
where hr denotes the hop count of route r.
h r = B(r, p )
pP
(4)
The total optical capacity equals to the total count of lightpath
T total =
T
pP r R
p
(5)
The total electronic processing traffic
T flows = r = t s
sS
(6)
Obviously,
T forward Ttotal T flows = T p t s
pP sS
(7)
We define virtual topology efficiency as
E=
T flows Ttotal
t = T
sS pP
s
(8)
p
Note that using (2) and (6), we have
E=
T flows Ttotal
= T
rR pP
r
p
1 H
(9),
where H is the traffic weighted average hop count,
B( r, p) H=
pP rR rR r
r
(10).
multi-hop WDM networks under dynamic data traffic. A conceptual model is accordingly proposed. Then the grooming issue is singled out and further addressed by ILP formulation and heuristic approaches. Two heuristic algorithms are analyzed through numerical results.
Therefore, E, H, or Ttotal, acts as one criterion of grooming performance for a certain virtual topology and routing algorithm. The higher the value of E, the lower the value of Tforward. 4. Heuristic Approach We now consider a heuristic algorithm to find the best virtual topology in terms of P and Tp under a random traffic demand set {t s}, as illustrated below. Algorithm A. (1) Suppose each s-d pair has a ligh...