A Pipelined Scheduling Strategy for Supporting Variable-Length Packets in WDM Networks

A Pipelined Scheduling Strategy for Supporting Variable-Length Packets
in WDM Networks

Shiann-Tsong Sheu, Yue-Ru Chuang and Hsuen-Wen Tseng
Department of Electrical Engineering, Tamkang University
Tamsui, Taipei Hsien, Taiwan 251, R.O.C.


Abstract: In single-hop star network with wavelength-division-multiplexing (WDM) technique, the packet/
wavelength assignment over a number of available channels is an important issue to provide low switching delay
and high bandwidth utilization. Basically, three pipelined stages are used to service packets: collecting packets
stage, scheduling packet stage and transmitting packet stage. Conventional scheduling algorithms only discuss
how to find the schedule with the minimal access delay for an individual group of variable length packets
regardless the pipeline feature. If the schedule of a group of packets cannot finish all transmissions before the
starting time of servicing the packets of the following group, some packets will suffer extra delay. In this paper,
we will propose a scheduling algorithm to overcome the potential problem by performing load balance in
channels. Simulations show that the proposed balancing approach indeed can reduce the average delay and
improve the load balance degree concurrently.

Keywords: single-hop star network, WDM, packet/wavelength assignment, pipeline feature, scheduling algorithm


1. INTRODUCTION
Wavelength division multiplexing (WDM) networks
offer a large transport capacity and are regarded as a
promising solution to the increasing demand of
bandwidth. A star-based single-hop WDM network
(SSWN) is a quite simple network topology, in which
all nodes connect to a passive star coupler (PSC). To
achieve the maximal network throughput, the SSWN
needs an efficient scheduling mechanism to reduce
processing time and average packet delay [1-6].
Most of the access protocols and algorithms can be
divided into two main classes. 1) preallocation-based
[7],[8], 2) reservation-based [9],[10] techniques.
Preallocation-based techniques use all channels of a
fiber to transmit data packets. These techniques assign
transmission rights to different nodes in a static and
predetermined manner. On the other hand,
reservation-based techniques particularly allocate a
channel as the control channel, to broadcast global
information about data packets of all nodes. As soon
as this information is broadcasted, all nodes invoke
the same scheduling algorithm to determine when to
transmit/receive a packet and on which data channel.
Since the reservation-based protocol outperforms the
preallocation-based protocol in most of situations, we
only focus our attention on the reservation-based
techniques in this paper.
In real cases, the way of serving packets in SSWN
includes three phases : collection phase, scheduling
phase and transmission phase. These phases are
executed in pipelining manner to smooth the traffic
flow. That is, in SSWN, after gathering the
information of arrival packets in the collection phase,
the scheduling phase is triggered; meanwhile, the next
collection phase starts. If some channels cannot
become free at the moment of serving the packets of
the next cycle, these packets scheduled in these
occupied channels will suffer extra delay. Therefore,
based on the nature of pipelining, the scheduler
should minimize the longest busy period among data
channels in a schedule cycle. This implies that the
scheduler with load balancing feature is more suitable
for such pipelined environment. Unfortunately, all
proposed scheduling algorithms in the literatures only
solve the best individual schedule for a given group of
packets in spite of the effect from pipelining. Hence,
it is desired to design an efficient pipelined scheduling
algorithm for SSWN.
There are two general policies often used for the
scheduler : longest-job-first (LJF) and shortest-job-
first (SJF). The LJF is a priority scheme and always
provides an advantage of balancing the loads in data
channels, but is known to result in relatively poor
average packet delay. The SJF, on the contrary, is also
a priority scheme that can reduce the average packet
delay by scarifying the feature of load balancing
1