Virtual Circuit Provisioning in WDM Networks

Bo Li and Yang Qin 
Department of Computer Science 
Hong Kong University of Science and Technology
Clear Water Bay, Kowloon
Hong Kong

    There exists a considerable distance in what today's optical
networks can provide and what the end users demand. For optical
network to hold greater promise in the future broadband networks, it
has to support wide range of applications with diverse service
requirements, exemplified by the rigorous efforts in the development
of both next generation integrated service Internet (IPng) and B-ISDN
ATM. The design of integrated service optical network poses a set of
challenging problems owing to the mismatches between the QoS
requirements of a variety of traffic types and the unique
characteristics of the optical networks, e.g., the variable bandwidth
requirement from end users and the fixed channel capacity provided by
the WDM network; more seriously, there are often conflicts in design
between the stringent QoS requirement of individual traffic stream and
the overall system performance such as throughput.
    In this talk we present our on-going work that explores the
possibility of providing the integrated service directly at the WDM
optical layer while taking into consideration the unique QoS
requirement from individual traffic stream. We consider what ARPA
wideband all-optical network (AON) refers as level-0 network based on
a passive optical network, which intakes the traffic from multiple
optical terminals (OTs). Each OT can serve as a feeder node for
multiple users.
    We introduce a framework based on a hierarchical link scheduling
mechanism. At the first level, packets or burst of packets are
scheduled from applications to the transmission queue of an OT; at the
second level, the actual packet transmissions are scheduled among
multiple OTs. We show such a separation is essential for achieving
scalability in that a potentially large number of user can be
accommodated, and for offering flexibility in that optimal scheduling
algorithm can be derived in each level best tuned to the specific
system requirements.
    The two key observations shaping the development of the link
scheduling algorithm are :

1. Multiple turnaround time can be justified for each individual
   traffic stream, since even the most stringent real-time traffic
   jitter requirement, e.g., HDTV video (1 ms) is still an order of
   magnitude larger than the end-to-end propagation time (25
   micro-seconds for a 5 Km span) and the tuning delay (typically in
   the range of a few micro-seconds) combined.

2. Packet transmission in each VC cannot be pipelined, as the
   in-sequence delivery of packets from VCs (real-time) mandates that
   each packet or burst must wait for the successful transmission of
   previous packet or burst.

    The preliminary results yield a number of interesting findings: 1)
the QoS guarantee for individual VC and high overall throughput can be
achieved; 2) the signaling overhead, including software processing,
end-to-end propagation delay and tuning time if any, can be
effectively masked out with the presence of reasonable number of VCs;
3) there is a trade-off between the complexity of the scheduling
algorithm and services it can offer.