Tải bản đầy đủ - 0 (trang)
Chapter 1. Orientation to Transport Networks and Technology

Chapter 1. Orientation to Transport Networks and Technology

Tải bản đầy đủ - 0trang

This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

of the important issues that distinguish transport networking from other far more dominant networking concepts. This establishes context

for the design problems that follow and equips a student to understand and explain to others—such as at a thesis defense—how it is, for

example, that we don't see individual packet routing considerations in a transport layer study, or how it can be that some nodes of the

actual network do not appear in the transport network graph. ("Does it mean you have decided not to provide service to those cities?")

These questions, both of which were earnestly posed by committee members to the authors students during their dissertations, are

typical of the misunderstandings about, or simple unawareness about, the concept of transport networking. Thus, there is a real need for

students and practitioners to understand transport as a networking paradigm of its own, different from telephony switching and packet

switching as well as leased-line private network design, and to be able to give clarifying answers to questions such as above. However,

space permits only a basic introduction to transport networking; the minimum needed to support later developments in the book. To delve

deeper, the book by K.Sato [Sato96] is recommended as a supplement. It gives a more extensive treatment of key ideas and

technologies for transport networking without considering network design or survivability.

[ Team LiB ]


This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

[ Team LiB ]

1.0.1 Aggregation of Service Layer Traffic into Transport Demands

Ultimately, whether through IP over optics or through a stack of DS-3, ATM and SONET layers, the net effect is that a set of user services

is mapped onto a set of physical high-capacity transmission, multiplexing and signal switching facilities that provide transmission paths to

support the logical connectivity and capacity requirements of all service layer flows. The aggregation of flows between each pair of nodes

on the transport network defines what is called the demand on the transport network (or the respective transport layer). The termdemand

has a specialized meaning, distinct from the more general term traffic [Wu92]. A demand unit is a quantum of transmission and routing

capacity used to serve any aggregations of traffic flow from the service layers. Whereas traffic may refer to measures of voice, data, or

video flow intensities (Erlangs, packets per second, Mb/s, frames/sec, etc.), demands on a transport network are specified in terms of the

number of managed units of transmission capacity that the aggregation of traffic requires. Demand may thus take on standard

transmission units such as lightpaths, OC-192s, OC-48s, DS3s or DS1s. An optical backbone network may typically be managed at the

OC-48 (~2.5 Gb/s of aggregated data) and the whole lightpath (i.e., contiguous optical carrier signal frequency assignments) level. Each

lightpath could be formatted to carry an OC-192 that may be structured as 4 OC-48s or to carry a 10GigE aggregation of Ethernet packet

frames, or many other application-specific payload formats for which mappings are defined. These are just examples of the general

concept of aggregations of payload being matched onto standard units of demand for routing and capacity management in the transport


Figure 1-1 illustrates the basic concept of multiple traffic sources being aggregated based on their common destination (or a route to a

common intermediate destination) thereby generating a demand requirement on the transport network. In the example, the bulk equivalent

of 76 STS-1s would in practice be likely to generate two OC-48 demands.

Figure 1-1. Traffic sources are aggregated from service layers into transport demand units

This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks.

It is important to note that, as the word "transport" suggests in general language, the individual packets, cells, phone calls, leased lines,

and so on are no longer recognized or individually processed at the nodes en route. In effect they have been grouped together in a

standard "container" for shipment toward their destination. Only the containers themselves are recognized and manipulated in the

transport network. The concept of containerization and transport, although commonly appreciated in trucking, shipping, air transport, mail

and overnight courier services, seems sometimes hard for those more familiar with packet switching or call switching to immediately

accept. The often strong presumption is of networks where every call or packet is inspected and routed at every node.


An example of a set of transport services that a carrier might offer, and the "bandwidth" that is allocated for each in transport capacity

planning is given in Table 1-1 where the basic unit of capacity planning is an STS-1. The table includes the traditional signal types such as

DS-1 through OC-48 as might be provisioned for private line services or for the operator's own inter-switch trunking requirements. Also

included are IP-based service offerings where the transport bandwidth allocated reflects the expected average utilization of the access

signal and the benefit of statistical multiplexing. For instance, an ISP with OC-12 interface links on its routers may request a "private line"

OC-12 as the bit-pipe to another router. Such a "PL" OC-12 will be a true OC-12 circuit for which 12 STS-1s are allocated. Alternately, it

may be an "IP" OC-12 service in which case a bandwidth of only about 13% of an actual OC-12 is allocated. The latter, obviously

lower-cost, service essentially just provides an OC-12 access interface, from which the IP payloads will be extracted and statistically

multiplexed with other flows in the carrier's network. Table 1-1 and Figure 1-1 are just examples. The general point is that in operation and

design of a transport network we deal with demand requirements posed in some basic unit of transmission capacity. These requirements

are generated by the aggregation of all types of service traffic. But in the transport network we generally never see or have to consider

individual packets or phone calls, etc., directly again.


A widespread practice is to use "bandwidth" and "capacity" as synonyms. Strictly, however,bandwidth is a

physical attribute of a transmission channel. It is the band of frequencies that the channel will pass below some

attenuation threshold, with units of Hz. The capacity of a channel is the rate of information that can be passed

through the channel under specific conditions of modulation, coding, power and noise, and has units of bits/s.


Table 1-1. Transport capacity allocated for various service types (STS-1 equivalents)

















OC-12 PL


IP-100T (Ethernet)


OC-48 PL


IP-GIGE (Ethernet)











PL = private line service in stipulated format, IP = IP packet service with specified interface rate and format, WL =

wavelength service bearing OC-48 or OC-192 container format.

In [DoHa98] the processes of bandwidth allocation based on service types, and overall of aggregation of point-to-point bandwidth

requirements to define a demand matrix for transport network design, are described further. That paper also describes an overall

framework of a network planning tool—the Integrated Network Design Tool (INDT)—that is relevant as an example of the kind of network

planning software tool where many of the design methods in the Part 2 chapters of this book would be employed in future.

This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

[ Team LiB ]


This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

[ Team LiB ]

1.0.2 Concept of Logical versus Physical Networks: Virtual Topology

Each fiber optic transmission system is itself an essentially fixed point-to-point structure that bears whatever set of tributary carrier signals

or wavelengths are presented to its inputs, up to its maximum capacity. Changes in the physical layer can be made, but on a much longer

time scale than required for purely logical reconfigurations in the transport network. The set of tributary signals (e.g., say STS3s) borne on

each fiber system (e.g., an OC-48) can be cross-connected to create a vast number of logical transport configurations for the higher

service layers. The pattern of logical point-to-point interconnection created by cross-connection of channels in the physical graph is also

called the virtual topology [RaSi98] [StBa99] [SiSu00]. The virtual topology has the same set of nodes but has an edge between each node

pair that have a path established between them. Figure 1-2 illustrates how the set of unit-capacity tributary signals at the input and output

of each essentially fixed point-to-point transmission system can be interconnected to provide a vast number of different logical transport

configurations. The different line thicknesses portray different amounts of point-to-point capacity between nodes. It is these virtual topology

configurations that different service layers perceive to be the physical transmission network.

Figure 1-2. A set of fixed point-to-point physical transmission systems and a small number of

the virtual networks that higher layer networks may be made to think is present.

The set of ways in which tributary channels on each fiber can be cross-connected to form different patterns of logical connectivity and

capacity is a combinatoric space. If the physical network consists of a set of spans S (indexed by j) with individual capacities Bj, then all

logical configurations that satisfy the following constraints are feasible:

Equation 1.1

This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

where Cx,y is the point-to-point capacity provided between nodes x and y in the logical configuration and

indicates the route over the

network taken to provide the x to y logical pipe:

means "for every...")

is 1 if the route crosses spanj, and zero otherwise. (

A frequent analogy for the reconfigurability of the transport network is the routing of trucks over a system of fixed highways with customer

payloads inside the containers carried by the trucks. This fits everyday experience to an extent but it belies the circuit-like nature of the

actual transport network. A system of point-to-point pipelines, each pipe having a finite flow limit, within which smaller rearrangeable tubes

are interconnected would be a more exact analogy. From the discussion, however, there are two important properties of the transport

network that are most relevant to the design problems that follow:

Many logical configurations of the transport network will be functionally indistinguishable by the service layers that use the

transport network. (This attribute we use for restoration.)

The same fixed physical transmission systems can be logically reconfigured to serve many different demand patterns. (This

attribute we use for traffic adaptation.)

These and further basic concepts about transport networking are illustrated in Figure 1-3 which is based on actual data for a European

inter-city transport network. Figure 1-3(a) shows the physical sets of transmission spans present. It is spans of this network that fail when

we postulate a cable cut. Each span in the physical graph (a) has an associated total capacity that is annotated on the figure and

represents the number of basic transmission channels or transmission layer links between each physically connected node pair. As shown

here, capacity is presented as (working, spare) couplet on each span. The working capacity supports the shortest-path routing of the

demands in (b). These are, for instance, individual lightwave channels on DWDM fibers between the nodes. Figure 1-3(b) shows the

pattern of logical interconnection ("demand") required in this network. In other words, Figure 1-3(b) is a graphical portrayal of what is

known as the demand matrix. The demand quantities are annotated on each dashed connection in (b) and would represent, for example,

the number of lightpaths required end-to-end between each node pair.

Figure 1-3. A physical transport network and overlying pattern of service layer demand.

Each unit of demand required between nodes in the logical layer has to be mapped onto a route over the physical graph and assigned a

transmission path that is a specific sequence of connected channels over the chosen route. The demand between N1 and N10 may be

served for example by paths over route (N1-N2-N3-N10) of the physical graph. But from the logical layer view, paths over route

(N1-N2-N7-N6-N10) would be equally as good (logically indistinguishable in fact) and thus if span (N2-N3) fails, these paths could be


This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

shifted from the first to the second route. However, the diagram makes it clear that if a span in (a) gets cut in isolation, there will not be an

isolated effect in (b). Many demand pairs will be affected by a single cut in (a). Thus, any such shifting over of the paths for node pair

(N1-N10) will have to be cognizant of the actual capacity on those other spans and coordinated with the corresponding switchover actions

for other affected node pairs that might also use capacity on those spans. In this regard, the spare capacity values shown in (a) are

reservations of additional capacity that is sufficient to support end-to-end path restoration (one of the schemes to follow) of all demands

whose working paths are disrupted by any single-span failure in (a). Conversely, Figure 1-3 lets us appreciate how the available capacity

on the edges of graph (a) can be cross-connected in different ways to support different patterns of demand in (b). This is also the

conceptual point about transport networks that Figure 1-2 is conveying.

The two network views, logical and physical, in Figure 1-3 also convey two basic classes of problem, depending on whether capacities or

demands are assumed as given quantities. One class of problem assumes that a forecast or planning view of the demand pattern is given.

This may actually be a family of possible future demand patterns, or, a stipulated "envelope" of maximum anticipated demands that the

network may have to serve. The problem then is to solve for the minimum cost allocation of transmission capacity in (a) (or addition of new

capacity to an existing set of capacities) that supports both the routing and protection of all demands in (b). Once a network is built,

however, the nearer-term operational problem is one of routing and configuring protection arrangement so that demands that actually

arrive are both served and protected within the current as-built capacity. Over the life of a network these two phases are revisited in a

constant cycle.

Unlike circuit switching for voice, the lifetime of connections in the transport network is generally much longer, typically days to years

because the aggregations of traffic do not change as rapidly as individual service connections do. For this reason, transport network

connections are often referred to as "semi-permanent" or "nailed-up" connections. The routing environment is also different from routing

through networks of trunk groups. First, in making rearrangements within the transport network (especially for restoration) there essentially

cannot be any blocking, because "blocking" in the transport domain means hard outage for all the services that the blocked carrier signal

would have borne. Secondly, once established, paths in the transport network may be very long-lived so there is much more impetus to try

to globally optimize the assignment of transport capacity. In contrast, the routing of any one call only commits the system for a few

minutes, after which time the system gets to start over with subsequent calls. In other words, the system state rapidly decorrelates in the

voice circuit-switched network (or the state of flows in an IP router-based network) but has a much longer correlation time in the transport


[ Team LiB ]


This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

[ Team LiB ]

1.0.3 Multiplexing and Switching

Multiplexing is the simultaneous transmission of many separate messages or lower-speed logical circuit connections over a shared


medium. Multiplexing can be via space, time, frequency or code division. Space refers to parallel physically distinct channels, such as

the separate fibers of a fiber optic cable or physically separate output ports on a switch or router. Time refers to synchronous time slot

allocation, as in SONET, or asynchronous time slot allocation as in ATM. Frequency division multiplexing (FDM) is the basis for

conventional AM/FM radio and the pre-fiber generation of high capacity analog and digital microwave transmission systems. When

applied to optical frequencies, FDM is the basis of both coarse and dense wavelength division multiplexing (WDM).


This section, through to Section 1.2.3, is adapted with permission from the tutorial portions of the Ph.D. thesis

by D. Morley [Morl01].

There are two basic types of switching that may be used in transport networks: packet switching and circuit switching. In a

packet-switched network, the information flowing from one node to another is broken down into sequences of packets at the sending

node before being transmitted over the network to the receiving node(s). When a packet arrives at a switching node, it waits in a queue

to be transmitted over the next transmission facility en route to its destination. At the receiving node, the packets are reassembled to

reconstruct the original information stream. Because the packets occupy the full capacity on a transmission facility only for their duration,

the transmission capacity can be shared over time with many other connections (or sessions). This is referred to as statistical

multiplexing. Statistical multiplexing takes advantage of the strong law of large numbers to obtain efficiency in bandwidth use. In this

context, the principle states that for a number of independent flows, the bandwidth necessary to satisfy the needs for all of the flows

together stays nearly constant, and much less than the sum of their individual peak rates, even though the amount of traffic in individual

flows can vary greatly. Intuitively, the averaging effect is easy to appreciate, especially when delay can be introduced through a buffer to

queue up the access to the transmission link. At any moment a few applications could be increasing their traffic while other applications

are reducing their traffic. The larger the number of sources, the more that individual uncorrelated changes balance each other out to

approximate a near-constant total bandwidth requirement. If a large amount of queuing delay is used, the constant bandwidth

approaches the overall average rate of all sources. In contrast with TDM multiplexing, the total bandwidth required is the sum of each

source's peak bandwidth. Packet switching is thus particularly well suited for data sessions that are characterized by short bursts of high

activity followed by long periods of inactivity. On the other hand a central issue with packet switching is that queuing delays at the

switching nodes are difficult to control and packet loss can occur from buffer overflow.

Under TDM each connection is allocated a given amount of transmission capacity (usually in both directions) between the origin and

destination nodes. Therefore, once the path (or circuit) has been established, the connection has a guaranteed transmission capacity

through the network for its entire duration. Unlike packet-switched networks, the capacity allocated to individual connections cannot be

used by other connections during inactive periods. The sum of the capacity allocated to all paths on a given transmission facility cannot

exceed its total capacity. Thus, if a transmission facility is fully allocated it cannot accommodate any new connections. If no other paths

with the required capacity can be found through the network, a new connection request must be rejected or blocked. In contrast, an

overload situation in a packet-switched network results in increased queuing delays and potential loss of data due to buffer overflows.

[ Team LiB ]


This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

[ Team LiB ]

1.0.4 Concept of Transparency

Service layer networks also differ from transport networks in that they are usually designed for a specific type of service and typically

contain user signaling and control functions for setting up and tearing down calls or connections. In contrast, transport networks provide

bulk transmission of aggregate information streams independent of the type of end-user services supported. For this reason, they are

sometimes referred to as "backbone" networks. One of the ideals of transport networking is that paths in the transport layer would be

completely transparent to the rate or format of payload applied to them. In other words, if a transport path was established from A to B,

users could put any form of payload they wanted on the path without encountering technical problems such as synchronization failure or

high error rates. Conceptually the ideal transport network would thus provide nothing more than "ideal wires"—on demand—to its service

layer clients.

As conceptually simple as the idea of a wire is, it is actually quite difficult to even approximate a lossless, constant-delay, infinite

bandwidth path that would support any payload signal format at all. This ideal was not achieved in the plesiochronous digital hierarchy

(PDH, i.e., pre-SONET), where only completely stipulated tributary formats (DS-1, DS-2, etc.) could be carried. In SONET, transparency

was somewhat more closely approximated in that a variety of payload mappings were defined to adapt non-traditional payloads, such as

an FDDI or Ethernet LAN signal, into the payload envelope of suitable high-rate SONET OC-n signals. The ideal is closer to being

realized with optical networking, where an almost lossless and otherwise low-distortion, near constant-delay, optical path approximates

an ideal "wire," up to a certain distance, onto which almost any form of payload can be applied. The distance limit is fundamental

because we cannot indefinitely preserve all attributes (phase, frequency, waveshape, amplitude, polarization, and most important of all,

signal-to-noise ratio etc.) of an analog signal as it is transmitted over an increasing length of fiber and number of optical amplifiers and

possible wavelength-changing transponders. Thus, a more practical notion is that of digital transparency [Dixi03], p.105 where any

format or rate of digital payload signal, up to some maximum working bit rate is accommodated. Such digital transparency is being

provided by recent developments such as digital wrapper and GFP, that we discuss further in Chapter 2.

[ Team LiB ]


This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks

[ Team LiB ]

1.0.5 Layering and Partitioning

Over many different multiplexing and transmission technologies, the basic routing and switching functions that they employ are logically

equivalent. The main difference is the unit for allocating capacity. In SONET the basic unit of allocation is an STS-n channel, whereas in

optical networking it is a wavelength (or to be precise, an optical channel with a specified bandwidth in Hz). There are other important

differences but from a functional perspective, these networking technologies can be modeled in a generic fashion for many purposes using

the same abstract concepts. This has several implications. First, because the basic functions are equivalent in nature, the same types of

network elements and architectures are implemented across the range of technologies. For example, the survivable ring architectures first

implemented in SONET can also be implemented in the optical network layer. Similarly electronic digital cross-connects (DCS) for SONET

remain logically equivalent in many regards to optical cross-connects (OXC) in the optical network layer. The adjacent layers in the

network form client-server relationships, in which transport layers perform signal multiplexing, transport and routing for one or more client

layers. For example, the SONET/SDH layer can accept payloads directly from either the PDH, ATM or IP layers. In turn, the resource

requirements from the SONET layer become payloads for the WDM layer. Figure 1-4 shows most of the currently possible inter-layering

transport relationships.

Figure 1-4. Examples of possible client/server associations in a layered transport network.

Each network layer can also be partitioned horizontally into tiers or subnetworks according to geographic and/or administrative boundaries,

as in Figure 1-5. This further facilitates design and operation and allows for different survivability schemes to operate autonomously by

separation in space. In practice, this partitioning usually reflects the differences in demand distribution, cost structures and topological

layout within a network layer. For example, it is common practice to partition a network layer into separate access, metropolitan inter-office

(or metro) and core (or long-haul) subnetworks. In an access subnetwork, most demands originate at remote switching offices and

customer premises and terminate back at a main switching office (or hub). A metro subnetwork connects main switching offices (or other

points of concentration) within a metropolitan area and demands are typically more uniformly distributed. Because span distances in

access and metro subnetworks are typically less than 25 to 50 km, nodal equipment costs (e.g., ADMs, DCSs) usually dominate total

network costs. Long-haul subnetworks, on the other hand, usually connect metropolitan areas on a national or international scale. Span

distances are much greater and distance-related costs for cable installation, amplifiers and regenerators can dominate the total cost.

Figure 1-5. Partitioned view of a transport network.

This document was created by an unregistered ChmMagic, please go to http://www.bisenter.com to register it. Thanks.

[ Team LiB ]

Tài liệu bạn tìm kiếm đã sẵn sàng tải về

Chapter 1. Orientation to Transport Networks and Technology

Tải bản đầy đủ ngay(0 tr)