Performance and Control of Integrated Hybrid Optical Networks
Abstract
In the last two decades optical technologies have progressed both in terms of
network layer functionality and coverage: evolving from transmission to switching,
and disseminating in all network segments. The catalyst is, in part, the need to
meet the bandwidth demands of an exponentially increasing amount of traffic
conveyed through all networks. In this context, optics are attractive with the high
bandwidth-distance product, low processing overhead, and low environmental
footprint. At the same time, innovations in ICT and the penetration of Internet
have not only affected the amount of generated data traffic but also created an
extensive range of applications with as extensive quality of service requirements,
from bandwidth hungry to time critical services. A competitive market challenges
carriers to deliver these services effectively at a low cost. Thus it is essential to
maximize the utilization efficiency of network resources and offer a wide variety of
differentiated services, i.e. be able to support the quality of service (QoS) needed by
the most demanding applications.
It is in this context that the integrated hybrid optical network (IHON) is propositioned
as an attractive solution for next generation optical networks. It enables
networks with high throughput efficiency of packet switching and strict service
guarantees of circuit switching. The unique property of IHON is that it combines
advantages from the best of both worlds, i.e. circuit and packet switching, in a
fully integrated manner, at the link level. Each switching paradigm marks one
class of quality of service, which can then be segregated further within the class.
The circuit-switching performance characteristics are preserved by the guaranteed
service transport (GST) class: fixed end-to-end delay, theoretically no packet delay
variation and no packet loss. This equips carriers with a premiumquality service,
e.g. as legacy technologies like time-division multiplexing and leased line. High
resource utilization is gained from the statistically multiplexed (SM) service class of
packet switching traffic. This empowers a flexible, cost-efficient and future proof
network that can serve current and future classes of services and maximize resource
function.
This PhD thesis builds upon previous work on IHON node architectures, pro posed and studied at theDepartment of Telematics, NTNU, and the commercialized
IHON architecture from TransPacket, Fusion H1. The first architecture is the OpticalMigration
Capable with service guarantees (OpMiGua). It was presented by S.
Bjornstad in 2003 and further studied in the PhD work of A. Kimsas in 2011. It is
the basis for the two other architectures that build upon its principle of multiplexing
circuit and packet traffic on the same wavelength links in a time interleaved
method. The second architecture, the three level integrated hybrid optical network
(3LiHON) architecture, proposed by N. Stol in 2010, enhances OpMiGua by adding
a real-time packet service. The third architecture, the commercial Fusion node
from Transpacket, implements the OpMiGua principle in Ethernet-based networks.
These three architectures are the basis throughout this work and the termIHON
was introduced to denote all architectures which implement the integrated principle.
Note that, in the IHON context, the circuit traffic is also packet traffic which is
switched in the network as in circuit-switching.
The research objectives are categorized into studies and enhancements of the
data plane mechanisms and control planes for IHON. The OpMiGua architecture is
considered throughout all the publishedwork, while 3LiHON and Fusion on specific
works. Part I presents the reasons behind hybrid optical networks, an overview
on related work, and the edge that integrated hybrids have on this category. The
contributions are then presented into two categories: data plane and control plane
studies for IHON.
In the data plane, three mechanisms are investigated in line with the objective
ofmaximizing the wavelength utilization: a time-window scheduler with electronic
buffers, buffer management techniques, and rate-adaptive segmentation. In relation
to previous work on OpMiGua, there are two strategic different choices. First,
the SM traffic is buffered electronically, and second, the system under investigation
has only one output wavelength. The electronic buffering enables more involved
mechanisms, which are further investigated without the influence of thewavelength
contention resolution domain, thus focusing on fine tuning their performance. The
proposed schedulingmechanism monitors the gaps between GST packets through
the time-window and finds, in the buffer withmultiple queues, SMpackets of suitable
size that fit in the gaps. Results show that it efficiently increases the utilization
of the available capacity up to 90% total throughput without loss. The scheduler is
ever more important for the demanding case of the left over capacity for serving SM
being highly fragmented, i.e. GST is not burstified at the edge node before entering
the IHON domain but has the same packet length distribution as SM. Furthermore,
this packetized traffic pattern is an important case as it can describe the traffic
pattern of a time-sensitive class of service that could require a premium service
like GST; it is thoroughly investigated in this PhD work as compared to previous
studies where GST is mainly burstified. Results show that it is the worst scenario
for the SM performance as the system saturates at lower loads, e.g. 20% lower load
as compared to a short burst of ten GST packets. This is because the wavelength
blocking probability for the SM class increases when the GST traffic consists of
single packets because of the wavelength reservation to GST. Therefore, enhanced scheduling techniques in combination with buffer management techniques are
especially important for cases of high GST packet traffic load in order to increase the
resource utilization and make use of gaps shorter than the time-window. Moreover,
it motivates the investigation of queuemanagement schemes and proposing the
mechanism of adapting the segmentation of SM packets on the GST rate for limiting
the buffer overflow and SMdelay performance. Results from simulations show that
just by increasing the number of queues from one to four the wavelength utilization
increases by 5%; segmentation, performed only based on the probability that the
systemmight saturate to avoid over processing, reduces the SMpacket loss ratio
66% and keeps the SM delay close to the buffering bounds.
A highlight of this PhD work are the first experimental demonstrations of an
integrated hybrid optical network through the Fusion prototype nodes. Theoretical
and simulation results on the scheduling and buffer management techniques are
then compared to the implementation results, showing good compatibility and the
feasibility of the system. It is verified that GST is carried with no packet loss and
ultra-low delay and delay variation, enabling transport of the most time-demanding
services, like time-sensitive traffic and packet layer synchronization information.
Field-trials through metro and long-haul networks are also conducted with synthesized
traffic and also real production traffic, carried as GST in the carrier network
of UNINETT. The high throughput efficiency is demonstrated by adding SM traffic
on the common wavelength links, doubling the average link utilization without
affecting- and transparently to- GST. To stress the system to its maximum, measurements
are also performed during 24 hours and seven day periods, containing
high GST traffic load up to 0.8 and running the link into saturation. The amount
of SM traffic that is added ismonitored, resulting in a mean of 48% increase of the
link utilization. Strong dependence on time of day is observed, underlining the
importance of dimensioning IHON and load balancing/protection for SM, which
are areas outlined for further work. A simple analyticalmodel is proposed for the
leftover capacity. Its results are compared with the gathered experimental results,
showing that the model gives a lower bound on the maximumSM inserted in the
network as a function of the carried GST load. This theoretical estimate is offered
as a guideline when dimensioning the network. Furthermore, it quantifies the costefficiency
of IHON with the higher throughput efficiency as compared to parallel
hybrids.
The second part of this thesis investigates control plane solutions and their
compatibility with IHON. The specific requirements that IHON poses to the control
plane are defined and mapped to the current generalized multiprotocol label
switching (GMPLS) standards, showing that a fully compatible IHON control plane
is feasible. The control framework itself is integrated, i.e. it is considered that one
single GMPLS instance controls both packet-switched and circuit-switched virtual
network topologies. The framework requires no changes to the existing GMPLS
architecture; thus, it enables the addition of the GMPLS control plane as an added
functionality offered by the integrated hybrid networks. Furthermore, the recently
proposed software defined networking (SDN) paradigm is investigated both for its compatibility in controlling IHON and, more generally, for the scalability in optical
networks. This is the first work of this type and sheds light into possible issues
that might arise. Numerical results characterize the limitations in network dimensioning
when considering an SDN controller implementation in the presence of
different flow mixes in combination with the high bandwidth of the optical domain.
Employing flow aggregation and/or parallel distributed controllers is outlined as
potential solution to achieve SDN network scalability.