Chapter 3.7 - Quality Of Service Support

3.7 QUALITY OF SERVICE SUPPORT

Service differentiation is viewed as an important requirement in future networks
which optical packet switches are expected to support. This includes support for
multiple classes of service that may require loss and/or delay guarantees. In
electronic switches, service differentiation is typically implemented by scheduling
algorithms that utilize the switch’s random access memory to enable a high-priority
packet to receive better treatment (in terms of delay and loss) than low-priority
ones. This is accomplished by allowing higher priority packets to preempt other
packets and be served ahead of them. Although a similar approach can be applied
to the optical packet switching network, it is recognized that the use of FDLs as
buffers will probably not yield the required flexibility or buffer depth. Therefore,
alternative approaches might be preferable.

Quality of service guarantees can be supported either by allocating network
resources according to class priority or by differentiated contention resolution [46,
52]. In the former approach, restrictions are placed on the usage of resources
such as wavelength converters and FDLs in accordance with class priorities. In
the latter approach, packet priorities are taken into account in contention scenarios
and high-priority packets are favored over low-priority ones.

Consider, for example, an OPS node with shared wavelength converters and three
service classes with high, medium, and low priority. In such a node, an incoming
packet that finds its wavelength busy on the desired output port can utilize one of the
shared converters in order to be shifted to an available wavelength. If all wavelengths
at the particular output port are busy or the packet cannot find an available
converter, it will be dropped. Obviously, the probability that the packet will find an
available converter increases with the number of accessible converters. Therefore,
differentiated treatment in terms of loss probability between classes can be achieved
if limits are placed on the number of wavelength converters that each class can
access. One approach is to partition the converters in sets such that a number of converters
are accessible by all service classes, some are accessible by the two classes
with higher priority, and some are used exclusively by packets with the highest priority
[46]. As a result, packets of high (or medium) priority have a better chance of
locating an available wavelength converter, which means that fewer losses are
expected to occur.

Another approach would be to use two-level differentiation taking into account
both the sharing of wavelength converters and the occupancy of the FDL buffer.
In such a scheme [46], a percentage of the available converters are shared by all
traffic classes and the rest are used exclusively by packets with the highest
priority. In order to differentiate between packets of the lower two classes, a
buffer occupancy threshold is introduced. Thus, if the buffer occupation is above
a certain threshold, the lowest priority packets are discarded, and medium- and
high-priority packets are accepted until the buffer is full. In this case, service differentiation
is achieved in the wavelength domain between the two classes with
the highest priorities and in the time domain between the two classes with the
lowest priority [2].

Apart from restricting the access to resources according to service class priorities,
one could consider limiting the efforts the node makes to find an available wavelength
that can accommodate an incoming packet. Consider, for example, a case
where all output wavelengths are busy and a packet has to be delayed until a wavelength
becomes available. Service differentiation can be achieved in this scenario if
the number of searches performed depends on the packet priority. Thus, the search
for a wavelength that can accommodate a high-priority packet can be performed
iteratively (for increasing delay values) in order to avoid packet dropping, and
fewer searches can be performed for medium-priority traffic, and only one search
for low-priority traffic [46]. Alternatively, if deflection routing is employed for contention
resolution, high-priority traffic may have more or better (shorter) deflection
paths than lower priority traffic in order to guarantee a lower loss probability and/or
a smaller delay.

< Previous Excerpt Next Excerpt >

Purchase This Book

TABLE OF CONTENTS

Chapter 3.7 - Quality Of Service Support

Contact Preferences

This is embarrasing...

Customize Your GlobalSpec Experience

Select Your Free Newsletters

Industry Newsletters

Select Your Free Product Alerts

This is embarrasing...