8/7/2019 Backbone 1
1/76
Seven
WAN Links and Backbone Protocols
Common Carrier Services for Building a WAN
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
8/7/2019 Backbone 1
2/76
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-ii
8/7/2019 Backbone 1
3/76
Objectives
Participants will be able to understand and explain the choices available to a network designer when selecting a wide areanetwork backbone technology. In particular, participants will be able to compare and contrast the relative merits of
conventional T1/T3 private line networks with newer packet and cell-based switched networks such as Frame Relay, ATM,
and SMDS.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-1
8/7/2019 Backbone 1
4/76
Glue for Wide Area Networks
Communications services provided by local telephone companies and common carriers form the backbone of wide areanetworks, assuring that packets can move as easily around the globe as they do between floors of an office building. The
services include the private lines (the oldest and most commonly used service), ranging from 2400 bps to 45 mbps, ISDN,
X.25, Frame Relay, SMDS, and ATM (the newest and least used service).
From almost no choice
Until the late 1980s, choosing the right network service was pretty easy. The network designer only had two choices. For high
bandwidth and fast response time, private line networks were the only choice. On the other hand, connecting hundreds of
locations together with private lines is a pricey undertaking, leading to the other choiceX.25 packet switched networks which
provide fault tolerant, reliable communications by means of a shared network cloud which transparently delivers packets
from any host to any host.
To Let me count the ways. . .
The network designer of the mid-1990s has several new choices in selecting wide area connectivity tools. In a note of
serendipity, the user demand for faster speeds1 has been met by a telecommunications industry rushing to deliver newer, faster,
and cheaper services that can glue WANs together.
Its an interesting confluence of needs and availability. During the 1980s, the telecommunications industry was planning new,
high speed switching fabrics that would be able to switch data transmissions, video signals, and electronic images as well as
ordinary telephone calls. At about the same time, the seeds were being planted for the new applications that would demand
high bandwidth, many-to-many connectivity.
1 Driven by newer applications that use more and more bandwidth to transmit information between pairs of computers. Examples include image
files, multimedia (for instance, desktop-to-desktop video and voice through the computer), and even simple files transfers, as users discover
volumes of information stored on distant computers that were once considered inaccessible.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-2
8/7/2019 Backbone 1
5/76
Glue for Wide Area Networks
Communications services form the backbone of the WAN
Private lines and X.25the only choices up through the 1980s
Late 1990s present new choices
Frame Relay, ATM, SMDS, & ISDN
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-3
8/7/2019 Backbone 1
6/76
Private Line Services
In Chapter 4, we talked about several private line service offerings, generally available from both local telephone companiesand interexchange carriers. Before considering Frame Relay, ATM, or SMDS, it may be helpful to take a closer look at some
of the underlying technology used to provide dedicated T1/E1 and T3/E3 (and fractional) services.
Getting from here to thereThe central offices of both local telephone companies and interexchange carriers have two major sub-systems, developed
primarily for routing and delivering ordinary telephone calls. Switching systems provide circuit switching (for telephone
calls), setting up temporary paths through the telephone network for the duration of a single telephone call; some switching
systems, however, provide packetor cellswitching (initially for data transmission, but soon for multimedia connections as
well), setting up temporary paths through the telephone network for each packet or cell.2
Transmission systems are the fiber optic, microwave, and satellite systems used to connect central offices to one another, andto customer sites as well.
How private lines are builtIn the example on the facing page, a router in San Francisco is connected to another in New York City. A private line provides
the connectivity, using the transmission systems of three telecommunications service providers. Note that the switching
systems, present in all central offices, do not participate in this connection. Rather, technicians for each of the providers make
permanent connections in each of the central offices, linking one transmission system to the next. The local loops, connecting
the customer sites to the nearest central office, are also hard-wired into to the transmission systems.
2 The term cell emerged in the 1980s as the unit of data which is moved through the network. The cell has a standard, fixed length of 53 bytes (or
octets), including 48 bytes for user information (called the payload), and 5 bytes for addressing and other overhead. If you think of the cell as a
small packet of fixed length, you wont be far off.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-4
8/7/2019 Backbone 1
7/76
Private Line Services
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-5
8/7/2019 Backbone 1
8/76
Private Line ServicesPrivate lines provide point-to-point links between pairs of routers, affording the simplest, most direct means of providing
connection.
24 x 7 service
When you order private line services, you have the full use of the link, essentially guaranteed 24 hours per day, 7 day per
week. If you order T1 service, you can be sure that your routers can transmit to each other at the full 1.536 mbps, hour after
hour, day after day, forever. Traffic from other companies or government agencies (using the same carrier) will never impact
your throughput (and yours will never impact theirs).
Typical speeds
56/64 kbps lines, T1/E1 lines (1.536/2.048 mbps), and fractional T1/E1 lines (N x 64 kbps) are all examples of private lines.For higher speeds, most carriers offer T3/E3 lines (45/34 mbps) as well as fractional variants, or even OC3 (155 mbps) or
OC12 (622 mbps).
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-6
8/7/2019 Backbone 1
9/76
Private Line Service
Point to point links
24 x 7 service
Typical speeds are 56 kbps to 1.536/2.048 mbps (T1/E1)
Higher speeds T3/E3, OC3, and OC12 are available
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-7
8/7/2019 Backbone 1
10/76
Advantages of Private Lines
Traffic requirements drive network choicesThe most challenging requirement facing the design team of a wide area network is to provide enough throughput in the
network to deliver good performance for the users, without spending too much money. As the explanations and exercises in
earlier chapters have made clear, the task requires an understanding of the traffic sources and sinks that drive a network. Once
the team understands the behavior of traffic, it must select routers that will support the required level of activity, and provide
alternate routing in the event of a failure.
Zero packet lossPrivate lines do not discard packets. Ever! When congested, Frame Relay, ATM, and SMDS networks will discard packets,
relying on your routers or computers to retransmit.
Looking for predictable performanceIt is always much simpler for a designer to understand end-to-end performance of a network when all the components in the
network have fixed, perfectly predictable throughput limitations. Ethernet and Token Ring LANs offer this predictability.
Routers offer it too. Private line links between routers also offer the same, perfectly predictable performance. If a half T1 link
connects two routers, the designer knows that 768 kbps is always available.
Design process is easy (or easier)With a number of new services available and competing for the job of handling traffic between pairs of routers, designers will
continue to choose private lines in many cases because they are easy to understand and to implement. The newer services,
such as Frame Relay, ATM, and SMDS will all work, but additional layers of complexity to the design process because they
all rely on allocating a shared resource (the switching systems) among competing users (the data packets or cells).
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-8
8/7/2019 Backbone 1
11/76
Advantages of Private Lines
Cardinal rule of network design (part 2)
Zero packet loss
Designers want predictable performance in every component
Private lines make the design process easier
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-9
8/7/2019 Backbone 1
12/76
PPPThe Point to Point ProtocolIf the designer chooses private line links between routers, then he or she should also select a standards-based link protocol
designed to run over point to point links. The Point to Point Protocol, defined by RFC 1661, is an industry standard designedto provide logical connectivity between routing protocols in two neighboring routers.
The private line (and DS1 framing) provide the physical connectivity. PPP provides a standard method for transporting
multi-protocol datagrams over point-to-point links.3
Before PPP
In the absence of any industry standard prior to PPPs general adoption, router vendors designed their own HDLC link
protocols, making it impossible for the routers of two different companies to talk to each other over a point to point link.
What does PPP do?
There are three capabilities that PPP provides:
1. Manages the link. Provides logical connection between the two routers on start-up, terminates the connection if the
link should fail, and assure that the link performs properly.
2. Handles encapsulation. Multiplexes different network layer protocols simultaneously over the same link, in a
recognized manner, so that the peer router can determine which network layer packet is contained in each frame, and
deliver the contents to the appropriate routing software upon receipt.
3. Implements protocol-specific handling. Network control protocols, developed for each of the network layer protocols
being routed, manage specific details required for each protocol.
3 Quoted from the RFC.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-10
8/7/2019 Backbone 1
13/76
PPPThe Point to Point Protocol
Provides logical connectivity between two neighboring routers
Standards-based (RFC 1661)
PPP provides:
1. Link management
2. Multiprotocol encapsulation (i.e.multiplexing)
3. Protocol-specific handling
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-11
8/7/2019 Backbone 1
14/76
PPP Operation
PPP Frame
The Point to Point Protocol is based on standard HDLC framing, and has the format shown on the facing page. The fields are
defined as follows:
FLAG. The standard HDLC sequence, 01111110 (Hex 7E).
ADDRESS. PPP requires that the address be 11111111 (Hex FF).
CONTROL. All frames are unnumbered information type (UI).
PROTOCOL. Identifies the layer 3 protocol which is encapsulated in the PPP frame, using the values
defined in RFC 1340 for Assigned Numbers. For example, 0021 indicates IP data
packets, while 002B indicates Novell IPX data packets.
INFORMATION. Contains the link establishment and maintenance, and disconnect commands during set-
up or disconnect. Contains the encapsulated layer 3 data packets once logical connection
is established. The RFC specifies that the maximum transmission unit (MTU) for PPP
must be 1500 octets or greater.
FCS. Can be either 16 or 32 bit CRC.
PPP State Diagram
The purpose of PPP is to allow routers to open logical links with one another, and exchange data frames. The state diagram
illustrates the process, beginning with a dead link. PPP then attempts to Establish a link with the neighboring router, thenmust Authenticate itself to its neighbor, and finally begin the process of transferring network layer information (denoted by
the box labeled Network).
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-12
8/7/2019 Backbone 1
15/76
PPP Operation
PPP Frame
Flag Address Control Protocol Information FCS Flag
Number of Bytes 1 1 1 2 0 - 1500 2 or 4 1
PPP State Diagram
Link Dead Establish AuthenticateUP OPENED SUCCES
FAIL
CLOSING
FAIL
DOWN Terminate Network
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-13
8/7/2019 Backbone 1
16/76
Frame or Cell-based Switching Schemes
Although private lines offer the designer an easy solution, and provide reliable, predictable performance, a number ofrelatively recent schemes based on packet or cell switching have emerged to offer viable alternatives to the fixed bandwidth
and topology offered by private lines. These newer schemes offer the potential of significant savings (for long-haul transport
of messages) and offer a strikingly high degree of fault tolerance.4 They have in common the use of new switching systems,
located in the central offices of local telephone companies and interexchange carriers.
Connecting many to many
The telecommunications industry has built its well deserved reputation for extraordinary reliability on the simple premise that
intelligent switching systems located in central offices can connect any telephone in the world to any other, by the use of a
call-setup process at the beginning of each telephone call. Whats more, the architecture of telephone networks in developed
countries have evolved with multiple alternative routes between major switching centers, offering the strong likelihood that our
calls will go through on the first attempt nearly every time.
Whats different about frame or cell-based switching?
Different switching systems, for one thing. The switch fabric that works so well for telephone calls is not particularly suitable
for handling the kind of routing decisions required. Frame Relay, ATM, and SMDS all use some form of packet switchingthe switch fabric reads the addressing information contained in the header of every packet or cell and makes instantaneous
routing decisions for each chunk.5
4 We shouldnt forget one thats been around for a whileX.25.5 Not unlike the process going on in the router, except that Frame Relay, ATM, and SMDS are all Layer 2 switching technologies, while routers
switch at Layer 3.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-14
8/7/2019 Backbone 1
17/76
Frame or Cell-Based Switching Schemes
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-15
8/7/2019 Backbone 1
18/76
Frame or Cell-based Switching Schemes
The users perspective
The illustration on the previous page presents the carriers view of how traffic is delivered between a single pair of routers
when using packet or cell-based switching. Its important to see the physical topology and to compare it to the private line
solution, but it doesnt really help in understanding the effect that this new telecommunications infrastructure has on the design
of your wide area network. Its important to view the network from the perspective of the user.
Ubiquitous connectionsfrom anywhere to anywhereOn the facing page we see a new, and elegant network topology. Gone are the multiplicity of T1 links between city-pairs.
Gone are the requirements for multiple WAN ports in every router. The wide area network is an intelligent cloud thataccepts packets or cells from any number of routers, and delivers each to the correct destination at the far side of the cloud.
The need to plan for adequate bandwidth and redundancy is given over to the local telephone companies and the interexchange
carriers! Or so it would seem.
How savings derive from packet or cell-based switching
A major portion of the cost in a wide area network is the provisioning of theT1/E1 (or FT1/FE1) access channels required to
connect a router to the long-haul network of the interexchange carrier. If we can recall the diagrams in Chapter 5 illustrating
the full mesh network connecting four cities, we will see that each router requires three T1 ports (in the router itselfabout
$1,000 each), three DSUs (another $1,000 each), and three access channels (about $125 to $1,000depending on the distance
spanned by the Access Channel, and on the RBOCper month each). Now consider the diagram presented here. Although
every city has links to every other city, those links are managed invisibly, inside the cloud, by the telecommunications service
provider. Each router requires only one port into the cloud, and only one access channel.
It is through this scheme that substantial economies can be derived.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-16
8/7/2019 Backbone 1
19/76
Frame or Cell-based Switching Schemes
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-17
8/7/2019 Backbone 1
20/76
Design Concerns
Sharing the access channel
Although lower costs result from the ability to share a single access channel into the switched network cloud, obviously the
traffic to and from any one router must fit on a single access channel. The designer must now be able not only to estimate
the peak load to each of the other cities, but also to understand how all remote destinations, taken together, drive the router port
and access channel at peak load. Any margin of error (gained by using multiple access channels) formerly available to
overcome unforeseen traffic peaks in traditional point-to-point networks evaporates in this new scheme. The design team is
forced to do more careful data gathering and make more accurate estimates.
Congestion problems in switches and cell discard
The switches used in Frame Relay, ATM, and SMDS networks are engineered to carry a predicted peak cell arrival rate.Although service providers make every effort to model activity of large groups of customers accurately (and engineer their
switches accordingly), no model can be exactly correct under all conditions. Bursts will occur in data traffic that momentarily
exceed a switchs design limits, causing temporary congestion. Buffers can fill up, and cells will be discarded. Well designed
switched data networks regularly maintain packet discard rates of 1 x 10-6 or below.
If a single cell is discarded, applications, transport layer protocols, or the routers will discover the incomplete transmission, and
retransmit an entire packet (or PDU), not just the one missing cell. Of course, the retransmitted PDUs may compound the
congestion, creating the possibility of unstable performance.
Some variability of propagation time and latency
Propagation time (how long an electrical or optical signal takes getting from here to there) and latency (how long the network
elementsswitches, for examplespend processing a packet or cell) are invariant when private lines are used as links. Packet
and cell-based switching systems, such as Frame Relay, ATM, and SMDS will introduce variability in both propagation time
and latency, creating some mathematical uncertainty in the network design. The effects of this uncertainty are usually not
large, but can become sizable during periods of congestion, and should not be totally discounted.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-18
8/7/2019 Backbone 1
21/76
Latency and propagation delay in terrestrial-based transcontinental private line circuits here in North America is typically on
the order of 27 to 30 milliseconds (for each direction). End-to-end latency and propagation delay in switched data networks
(Frame Relay, ATM, and SMDS) can range between 45 to 120 msec in North America.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-19
8/7/2019 Backbone 1
22/76
Design Concerns
All traffic must share the single access line during peak traffic
Congestion problems in switches and cell discard 1 x 10-6
Some variability of propagation time and latency
45 to 120 msec
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-20
8/7/2019 Backbone 1
23/76
Congestion, Latency, and Cell Discard
A simplified view of a cell-based switch is shown on the opposite page. Cells are arriving from several input queues, and areall destined for the same output queue.
When everything works OK
Cells are buffered momentarily at each input queue (San Francisco, St. Louis, and Dallas ports) until the processor can move
them into the switching fabric. Once inside the switching fabric, each cells addressing information is read and compared with
a routing table that associates each address with an output queue (or port).
As long as the combined arrival rate of all cells destined for the same output queue is less than or equal to the speed of the
output port, cells are moved out in an orderly fashion, although almost all of them will have to wait at least a fraction of one
cell time (the time for one cell to be transmitted onto the wire) in the buffer
until its their turn to actually hop onto thewire.
Delay is introduced
Now imagine that cells arrive at the output queue (headed for Boston) faster than the speed of the port. The buffer begins to
hold the cells, sending them as rapidly as the wire permits. Each cell is delayed in this output queue. The amount of delay
depends on the speed differencehow much faster do cells arrive than they departas well as on the duration of the burst. The
switch is momentarily congested.
The bit bucket
If the burst goes on too long, and the output buffer fills up, cells will be discarded, not just delayed.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-21
8/7/2019 Backbone 1
24/76
Congestion, Latency, and Cell Discard
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-22
8/7/2019 Backbone 1
25/76
Frame Relay
Frame Relay sales have exploded since the mid-1990s, displacing many existing private line networks, and winning newbusiness as well. From a technical perspective, Frame Relay provides an excellent transport mechanism for linking LANs and
hosts across the country, or around the globe.
Whos a candidate for Frame Relay?The natural market for Frame Relay is the corporation, government agency, or institution with a number of LANs located in
different cities, states, or countries. LAN to LAN traffic and LAN to host traffic patterns are a good match for the operational
characteristics of Frame Relay. Having a number of locations (at least three or four), all of which need to exchange
information with each other, at widely separated distances is usually a good decision criterion for selecting Frame Relay.
Business Communications Review (July 1997) reported that users might find cost savings between 20% and 50%, when using
Frame Relay to build wide area networks, as compared to comparable private line solutions.Whos not a candidateFrame Relay can become expensive and is very difficult to implement and manage when the number of sites which must
communicate directly with each other gets large (perhaps 100 to 150 fully meshed locations). Furthermore, until recently,
Frame Relay switches havent included the ability to provide fractional or full T3 speeds T1 has been the highest speed
available. In these cases (very large networks or very high speeds), the implementer should consider ATM or SMDS as a
better choice.
From an operational perspective, applications which need real-time throughput (such as process control, video, or voice)
cannot be properly supported on Frame Relay, due to the inherent performance limitations of the protocol.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-23
8/7/2019 Backbone 1
26/76
Frame Relay
Accounts for as much as 1/3 of data networks
LAN to LAN connections and LAN to host connections are good applications
Frame Relay is not suitable for large, fully meshed networks, very high
speeds, or real-time applications
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-24
8/7/2019 Backbone 1
27/76
Pricing Frame Relay Ports and PVCsOn a national scale, all of the major carriers offer Frame Relay service. AT&T, CompuServe, Sprint, and WorldCom all offer
ubiquitous service throughout the U.S.6 In addition, most RBOCs, GTE and many larger independent telcos offer Frame Relay
for Intra-LATA service.
Varying price strategies and technical characteristics among carriersMost providers charge for two pricing elements:
1. The physical port attaching to their Frame Relay networks (i.e.FT1, T1, FT3 or T3 access channel). Each router or
host connecting to Frame Relay will require at least one port. Ports are available for access speeds ranging from 56
kbps to 45 mbps.7 The higher the access speed, the more expensive is the port.
2. The permanent virtual circuits (PVCswhich provide the logical connections) between pairs of routers or between
hosts and routers. Carriers configure permanent logical pathsthe PVCsbetween all pairs of devices which mustcommunicate directly when Frame Relay is installed. This is the means by which full mesh topology is provided inside
the cloud.
PVCs are also priced by speed, ranging from 0 kbps to fractional T3 speeds of as high as 34 mbps (some carriers also
add a distance element to the price of a PVC). It is here that the network designer must use care in estimating the
traffic between pairs of locations. Although any one PVC can transmit bursts up to the full speed of the port, the Frame
Relay service provider may drop frames that exceed a committed information rate (CIR)the speed of the PVCif
their network is too busy to handle the momentary overload.
6 Stentor and Unitel also offer the service in Canada.7 T3 port speed is still relatively new, and may not be available from all carriers or in all areas.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-25
8/7/2019 Backbone 1
28/76
Pricing Frame Relay Ports and PVCs
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-26
8/7/2019 Backbone 1
29/76
Pricing Comparison
In order to demonstrate the pricing concepts concretely, lets compute the costs for three alternatives to link a single pair oflocationsT1, fractional T1, and Frame Relay. In this example, we will use prices for AT&T services for all three
alternatives.
San Francisco to New York T1
We obtained the price for a T1 private line, including access channels on both ends, from the normal AT&T interstate tariffed
rates. Customers with special pricing plans will have different results.
T1 private line $14,334
San Francisco to New York FT1 (768 kbps)
We again obtained the price for a fractional T1 line, including access channels on both ends, from normal AT&T tariffed rates.T1 private line $11,469
San Francisco to New York Frame Relay
The prices reflect actual tariffed rates provided by AT&T; however, actual customer pricing for the Frame Relay ports and the
PVC tend to be somewhat flexible (unlike the access channels, which are largely driven by RBOC pricing policies). Look for
greater flexibility in access channel pricing in 1997, when local competition may tend to drive down these prices.
1 T1 access channel for San Francisco $525 525
1 T1 access channel for New York 886 886
2 1.544 mbps Frame Relay ports 2,468 4,9361 1.024 mbps PVC 2,227 2,227
TOTAL $8,574
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-27
8/7/2019 Backbone 1
30/76
Pricing Comparison
San Francisco to New York T1 $14,334
San Francisco to New York T1 $11,469
San Francisco to New YorkFrame Relay with 1.024 mbps PVC $8,574
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-28
8/7/2019 Backbone 1
31/76
Frame Relay Design Alternatives
Full Mesh
The Frame Relay network in the diagram at the left illustrates the full mesh design alternative. Each router has six PVCs, so
that it has a direct logical (not physical) connection with all other routers in the network. Although this design seems like a
reasonable choice at first glance, there are concerns in implementing it. A casual look reveals one issue. The number of PVCs
grows dramatically with the number of routers, increasing the cost and complexity of the network.
Another concern is a subtle problem having to do with to do with the way routing algorithms recognize resources in the router.
Generally, the routing algorithm sees the physical V.35 interface and Frame Relay protocol as a homogeneous resource.
Although individual packets are mapped to the correct PVC by the Frame Relay implementation in the router, the routing
algorithm itself is usually unaware of the granularity at the interface. There are work-arounds that overcome most of the
limitations, at the expense of added configuration effort by the router installation team.
Hub Router
The diagram at the right side of the page illustrates a common solution being implemented in large Frame Relay networks. In
this alternative, a high powered router sits at the center of the network, forwarding all traffic to the correct destination router.
Each outlying router uses a single PVC to forward packets toward the hub. Both problems described above are solved. The
number of PVCs is small, and, because theres only one PVC, the routing algorithm has full knowledge of the port
configuration.
If this network architecture is used with OSPF, IS-IS, or NLSP (link state protocols), the hub router requires special
configuration to manually configure network layer addresses for each PVC.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-29
8/7/2019 Backbone 1
32/76
Frame Relay Design Alternatives
Full Mesh Hub Router
( )Number of PVCs =
N N 1
2
where N is the number of sites
How many PVCs would be required for a full mesh if the
network consisted of 25 routers? 50 routers?
Number of PVCs = N
where N is the number of router sites excluding the
hub
How many PVCs would be required for a hubbed network
of 25 routers? 50 routers?
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-30
8/7/2019 Backbone 1
33/76
Router Unicasting and Frame Relay
Routers of course maintain a separate routing table and/or link state database for each protocol being routedTCP/IP,NetWare, AppleTalk, DECnet, and so on. As you will recall from Chapter 6, a key part of every routing algorithm is the
requirement for periodic update messages sent by each router to all peers to which it is directly linked. These update messages
provide automatic propagation of reachability information, and allow each router to construct a table or database, which tells
the routing protocol how to forward packets to all known networks and subnets. In addition, NetWare, AppleTalk, and VINES
networks involve regular timed broadcast packets to enable workstations to find and connect with various services.8
In the traditional point to point network, each WAN interface has only one peer router at the far end of a link. So a router with
three physical links to its peers sends regular routing update messages (or link state messages) to just one other router on each
link.
In a Frame Relay network, however,each router has many peers, and must send these updates and broadcasts to each. In thediagram, Router A sends a RIP packet (advertising all IP networks in city A) to Router B, then a second, identical packet is
sent to Router C, a third to D, and so on. The sequence is repeated every 30 seconds. But thats not allthe Novell IPX
networks are identified with Novell RIP packets which are also sent, in order, to each of the peer routers (usually at 60 second
intervals). Finally, Novell Service Advertising Protocol packets are sent to each of the peer routers (also at 60 second
intervals). Even though the contents of each RIP or SAP packet is identical, router A sends each one five times, varying only
the DLCI which uniquely identifies each router to the frame relay network. 9 The implementers of this network chose to use
the point-to-point paradigm in describing the frame relay cloud to the routers, failing to take advantage of an addressing
method known as multicasting.
8 Which are sent as router-to-router unicast packets across the WAN cloud.9 AppleTalk routers send RTMP packets (the equivalent of RIP) every 10 seconds.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-31
8/7/2019 Backbone 1
34/76
Router Unicasting and Frame Relay
F R A M E R E L A Y
N E T W O R K
N e t w o r k s e n d s e a c h p a c k e t to
a p p r o p r i a t e d e s t i n a t io n
R o u t e r A
R o u t e r B
R o u t e r C
R o u t e r D
R o u t e r ER o u t e r F
E v e r y 3 0 s e c o n d s :
R I P ( I P ) P a c k e t to R o u t e r B
R I P ( I P ) P a c k e t to R o u t e r C
R I P ( I P ) P a c k e t to R o u t e r D
R I P ( I P ) P a c k e t to R o u t e r E
R I P ( I P ) P a c k e t to R o u t e r F
E v e r y 6 0 s e c o n d s
R I P ( I P X ) P a c k e t t o R o u t e r B
R I P ( I P X ) P a c k e t t o R o u t e r C
R I P ( I P X ) P a c k e t t o R o u t e r D
R I P ( I P X ) P a c k e t t o R o u t e r ER I P ( I P X ) P a c k e t t o R o u t e r F
S A P P a c k e t ( s ) t o R o u t e r B
S A P P a c k e t ( s ) t o R o u t e r C
S A P P a c k e t ( s ) t o R o u t e r D
S A P P a c k e t ( s ) t o R o u t e r E
S A P P a c k e t ( s ) t o R o u t e r F
E v e r y 3 0 s e c o n d s , R I P ( IP ) P a c k e t f r o m R o u t e r A
a n d
E v e r y 6 0 s e c o n d s , R I P ( IP X ) a n d S A P P a c k e t s fr o m
R o u t e r A
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-32
8/7/2019 Backbone 1
35/76
Router Multicasting and Frame Relay
The diagram on the opposite page shows how the use of multicasting (sometimes called group addressing) reduces the numberof routing information and SAP packets sent into the cloud from Router A. The router sends a single copy of each packet into
the cloud, using a specially assigned DLCIthe multicast address (sometimes called a group address). The carriers switch is
pre-programmed to understand the multicast address, and send copies of the original packets to each router that was defined as
part of the group.
In order to make this architecture work, the implementation team must negotiate the group addresses and group members with
the carriers implementation team when ordering the Frame Relay network.
Unfortunately, multicasting is not widely available for Frame Relay at this time. 10 Most carriers deployed their Frame Relay
switching networks before the multicasting standard (Frame Relay Forum implementation agreement 7) had been adopted by
the Frame Relay Forum. Although most frame relay implementations support some scheme for multicasting, not all of themwill interoperate. To learn whether you will be able to use it with your frame relay network, first check your routers software
to see whether and how it supports the group addressing or multicasting over Frame Relay, 11 and then ask your carrier if their
switching architecture is compatible.
10 In a market survey conducted by Distributed Network Associates in January 1998, only one Frame Relay provider (of 29 surveyed) was currently
offering multicasting. See11 In Ciscos IOS router software, you can configure Frame Relay Broadcast Queue to take advantage of multicasting routing updates over
Frame Relay.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-33
8/7/2019 Backbone 1
36/76
Router Multicasting and Frame Relay
F R A M E R E L A Y
N E T W O R K
N e t w o r k r e p l ia t e s e a c h m u l t i c a s t p a c k e t , a n d s e n d s
a c o p y t o a l l m e m b e r s o f t h e m u l ti c a s t g r o u p .
R o u t e r A
R o u t e r B
R o u t e r C
R o u t e r D
R o u t e r ER o u t e r F
E v e r y 3 0 s e c o n d s :
R I P ( IP ) P a c k e t to R o u t e r B w i th
M u l t i c a s t D L C I
E v e r y 6 0 s e c o n d s
R I P ( I P X ) P a c k e t t o R o u t e r B w i th
M u l t i c a s t D L C I
S A P P a c k e t ( s ) t o R o u t e r B w i t h
M u l t i c a s t D L C I
E v e r y 3 0 s e c o n d s , R I P ( I P ) P a c k e t fr o m R o u t e r A
a n d
E v e r y 6 0 s e c o n d s , R I P ( I P X ) a n d S A P P a c k e t s f r o m
R o u t e r A
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-34
8/7/2019 Backbone 1
37/76
Frame Relay Protocol
The Frame Relay protocol is a data link layer protocol, unlike its older cousin, X.25, which is a network layer protocol.Although the basic concepts are the same, the execution of the details is much different. The essence of the differences can be
summed up in a few wordsFrame Relay is a simpler protocol, optimized for fast transmission over high quality digital
circuits which rarely introduce bit errors, whereas X.25 was designed to assure error-free transport in the face of older (and
noisy), error prone telephone circuits.
When something goes wrong, throw away frames
The sliding window concept (send n number of frames and stop to await an acknowledgment from the receiver before
continuing) does not exist in Frame Relay networks. Hence, there is no on-off valve which limits the amount of traffic a
router might send toward the Frame Relay network. Instead the network is entitled to throw away frames that exceed an
agreed upon transmission ratethe CIR. If the router is transmitting faster than the CIR, the network may throw frames awayrandomly. By setting the DE (Discard Eligible) bit in selected frames, the router can tell the network which frames are
expendable (this feature is not widely used). Ciscos IOS allows network managers to determine classes of traffic in which the
DE bit can be set. Frame Relay switches will also throw away frames that have errored frame check sequences, or that are
otherwise corrupted.
DLCI An addressing schemeThe data link connection identifier is a virtual circuit number, assigned by the Frame Relay provider, that is mapped inside the
network to the PVC that connects specific pairs of routers. The DLCI only has meaning in the local connection between the
router and the nearby Frame Relay switch.
RFC1490 Identifying the network layer protocolThe Internet community has provided a standardized method of identifying to the receiving router which network layer
protocol is contained in each frame (provides a similar function to the protocol field in the PPP), and whether the Frame
Relay frame contains a bridged frame or routed packet. Unique numbers are assigned to each network layer protocol by ISO
(using NLPID) or IEEE (using SNAP).
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-35
8/7/2019 Backbone 1
38/76
Frame Relay Protocol
Flag Address Information FCS Flag
DLCI CR EA DLCI FECNBECN DE EA
Bit
Byte
8 7 6 5 4 3 2 1 8 7 6 5 4 3 2 1
1 2
Control NLPID or SNAP
NOTES:
1. NLPID is the ISO Network Layer Protocol Identifier.2. SNAP is the IEEE Subnetwork Access Protocol.3. RFC1490 defines how the NLPID or SNAP field is encoded for bridges and routers.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-36
8/7/2019 Backbone 1
39/76
Frame Relay Local Management Interface
The Missing Piece
The initial Frame Relay specification did not include any method for the router to exchange management messages with thenetwork providers switch. Vendors and service providers quickly recognized that routing algorithms, unless made aware of a
Frame Relay network outage, would be slow to update their routing tables. A means to verify the integrity of the Frame Relay
network on a continual basis was needed. In addition, they wanted to provide an automated way for the router to learn the
DLCI numbers assigned by the carrier. Two similar schemes emerged to provide a solution. More recently, a third local
management interface scheme was introduced by the ITU-T, co-incident with the definition of switched virtual circuits.
The original scheme Ciscos LMIAn ad hoc industry group published the Frame Relay Specification with Extensions in 1990, which contains the LMI
specification.12 It provides a regular exchange of link integrity messages between the switch and the router, and a means of
updating the routers list of PVCs (and their DLCIs). In Ciscos IOS software, this method is unabashedly referred to ascisco.
ANSIANSI standard T1.617-1991 includes Annex D (an addendum to the underlying standard) which specifies link integrity
messages and a means to update the routers list of PVCs. The functionality is nearly identical with that provided in LMI.
Differences not significantLMI and Annex D interface management differ only in the bit patterns and fields assigned for similar functions. Either can be
used equally well; carrier and router vendor should be consulted for compatibility. Among the differences is the assignment of
DLCI 0 for signaling under Annex D, while LMI uses DLCI 1023 for the same purpose.
12 The group included Northern Telecom, Stratacom, Cisco, and Digital Equipment
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-37
8/7/2019 Backbone 1
40/76
Frame Relay Interface Management
The Missing Piece
The Original, Ciscos LMI
ANSI T1.617, Annex D
Differences not significant
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-38
8/7/2019 Backbone 1
41/76
ATMAsynchronous Transfer Mode
This newest of the high speed backbone technologies is also getting the most attention in the industry, from equipmentmanufacturers to trade journals to users. The interest stems from the growing view that ATM will, in the future, form the
infrastructure for a new type of telephone networkone which is specifically designed to provide multimedia communications.
In addition, adapter cards and premises ATM switches (that is, non-telco switches) are widely available for building an ATM
LAN infrastructure allowing ATM to be not just the backbone protocol, but the protocol linking workstation to workstation, no
matter whether theyre in the same building or around the globe.
What it does
Think of ATM as a replacement for Ethernet or Token Ring; it is an access method for allocating bandwidth among competing
users. However, where Ethernet and Token Ring are technologies suitable only for LANs, ATM can work equally well in both
the local and wide area paradigms. The ATM switching systems act as both hub and router (albeit it Layer 2 routing) device
all in one. Architecturally, it makes no difference whether the switch is in a wiring closet of an office building, or in the
central office of a local telephone company or interexchange carrier.
Why is it so important?
Carriers world-wide are busy building out their transport networks with ATM as a backbone technology for data transmission,
carrying Frame Relay, SMDS, and native ATM traffic over a common infrastructure. Consider the network diagram provided
by AT&T shown here.
Quality of ServiceUltimately, some industry pundits believe that ATM will be the common infrastructure for transport of voice, video, and data
ATM was designed with several so-called qualities of service, so that voice and video are allocated fixed portions of
bandwidth at the beginning of a call to guarantee real-time delivery, while data can be allocated bandwidth on demand,
matching its bursty nature.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-39
8/7/2019 Backbone 1
42/76
ATMAsynchronous Transfer Mode
Technology for new telecommunications infrastructure, and . . . for
LANs, too
An example of a carriers ATM infrastructure map
Qualities of Service real-time delivery for voice and video, bandwidthon demand for computer networking.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-40
8/7/2019 Backbone 1
43/76
ATM Asynchronous Transfer Mode
ATM is about cell-based switchingThe information units defined for ATM are 53 byte cells. 5 bytes are allocated to the header (which contains full routing
information for each cell) and 48 bytes to the payload (higher layer protocols and the users information).
How the switching works
Cells arriving at a switching system from many different devices will wait briefly in queue, each awaiting its own time slot. 13
Some cellsthose for uncompressed video and voicewill have pre-arranged time slots to guarantee on-time delivery; these
information types generate continuous streams of bits, and require real-time performance, or the person seeing (or hearing) the
received information will be aware of motion stopping briefly, then suddenly catching up (speech would seem to be herky-
jerky, stopping briefly, and then catching up in a rush of sounds). Other cellsthose for data transmissionwill be given as
much bandwidth as needed (up to the limits of the physical speed of the port) for brief periods perhaps hundreds ofmicroseconds to tens of milliseconds.
When a time slot matching the needs of the sender becomes available, the switch reads the routing instructions in the cell, and
sends it on its way. Thats all! The switch immediately moves on to processing the next cell.
Routing information
Two layers of routing information are carried in each cella virtual path number, identifying a link between a pair of end-
points, and a virtual channel number for the specific logical connection between those two end-points.
13 Each device has its own dedicated link to the shared switch.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-41
8/7/2019 Backbone 1
44/76
ATMAsynchronous Transfer Mode
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-42
8/7/2019 Backbone 1
45/76
Designing ATMPorts and PVCsOn a national scale, most of the major carriers offer ATM service. AT&T, GTE, MCI WorldCom, and Sprint all offer service
throughout the U.S. In addition, Ameritech, Bell Atlantic, BellSouth, Pacific Bell, Southwestern Bell, SNET, and U.S. Westall offer some form of public ATM services in their market areas.
Varying price strategies and technical characteristics among carriersMost providers charge for two pricing elements:
1. The physical port attaching to their ATM networks. Each router or host connecting to ATM will require at least one
port. Ports are available for DS1 (1.5 Mbps), DS3 (45 Mbps), and OC3 (155 Mbps). GTE currently offers OC12 (622
Mbps) as a regular part of their service offering, while AT&T and MCI WorldCom offer it on an individual case basis.
The higher the access speed, the more expensive is the port.
2. The permanent virtual circuits or switched virtual circuits (PVCs or SVCs
which provide the logical connections)between pairs of routers or between hosts and routers. Carriers configure permanent logical pathsthe PVCsbetween
all pairs of devices which must communicate directly when ATM is installed. This is the means by which full mesh
topology is provided inside the cloud.
PVCs and SVCs are also priced by speed, ranging from 1.5 Mbps to 155 Mbps (some carriers also add a distance
element to the price of a VC). It is here that the network designer must use care in estimating the traffic between pairs
of locations. Although any one VC can transmit bursts up to the full speed of the port, the ATM service provider may
drop frames that exceed a committed information rate (CIR)the speed of the VCif their network is too busy to
handle the momentary overload.
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-43
8/7/2019 Backbone 1
46/76
Designing ATM Ports and PVCs
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-44
8/7/2019 Backbone 1
47/76
ATM Quality of Service
The key idea of QoS is pictured on page 7-41
each end-point negotiates for the type of service it requires whenever aconnection is made, and then sends its cells to the switch, where they are queued for delivery when the next time slot becomes
available. Now lets examine how each QoS is meant to work.
Quality of service
There are four service levels or qualities of service. Because the ATM adaptation layer (or AAL) is often related to the QoS,
theres a danger of confusing the two. with the QoSas the most widely used way of specifying the different levels. Another
term, Class of Service, identified by letters A - D, has also been used to describe the type of ATM service.
Constant Bit Rate (CBR) is a connection-oriented service, and provides circuit emulation, and is intended primarily
for voice telephony and carrying conventional, uncompressed video.
Real Time Variable Bit Rate (Rt-VBR) is a connection-oriented service, and provides a guaranteed timing
relationship between sender and receiver even though the bit rate is not constant. It is intended primarily for
transmitting compressed voice or compressed video. This QoS is currently offered by only BellSouth and MCI
WorldCom.
Non-Real Time Variable Bit Rate (Nrt-VBR) is also a connection-oriented service, and is the QoS most often used
for data communications applications. This QoS most closely resembles Frame Relay, with a committed
information rate, and the ability to accommodate brief bursts over the CIR.
Available Bit Rate (ABR) is a connection-oriented service that could be used for some data communicationsapplications. Among the carriers, it is currently provided only by U.S. West and MCI WorldCom.
Unspecified Bit Rate (UBR) is a connection-oriented service, and is the most risky of the various QoS that have
been defined. There are no commitments by the carrier or customer on bit rates or performance. Providing
transport at low cost, UBR might be used for forwarding email, low-priority file transfers, and other applications in
which significant delays are not particularly important.
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-45
8/7/2019 Backbone 1
48/76
ATM Quality of Service
Constant Bit Rate (CBR) circuit emulation for uncompressed voiceand video
Real Time Variable Bit Rate (RtVBR) variable bit rate, guaranteedtiming, for compressed voice or video
Non-Real Time Variable Bit Rate (NrtVBR) variable bit rate, mostcommonly used for data communications
Available Bit Rate (ABR) used for data communications
Unspecified Bit Rate (UBR) used for low-priority datacommunications applications, such as email or file transfers
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-46
8/7/2019 Backbone 1
49/76
ATM Adaptation Layer AAL
ATM specifies several sub-layers within the OSI data link layer
the Convergence layer, the Segmentation and Reassemblylayer, and the ATM Adaptation layer. AAL specifies how voice, video, and data output are mapped from the OSI network
layer into the ATM 53 octet cell structure.
For example, AAL1 specifies how Constant Bit Rate services such as real time voice and video are mapped into cells, while
AAL2 was specified for mapping Real Time Variable Bit Rate services, such as compressed voice and video.
The ATM adaptation layer for data users went through a number of fits and starts, before the ATM Forum adopted two
strategiesAAL3/4 and AAL5.14 Of the two, almost all vendors and carriers are choosing to offer Non-Real Time Variable
Bit Rate service over AAL5.15
14 Initially, there were separate, distinct AAL3 and AAL4 definitions. The members of the ATM Forum eventually found the differences to be
small, and the single AAL3/4 was the result.15 AAL3/4 is used for a single applicatonproviding SMDS over an ATM infrastructure. See page 7-70 for protocol details.
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-47
8/7/2019 Backbone 1
50/76
ATM Adaptation Layer AAL
Specifies how output from network layer is mapped into 53 octet cells
AAL1 used for Constant Bit Rate QoS
AAL2 used for Real Time Variable Bit Rate QoS
AAL3/4 and AAL5 used for data services (Non-Real Time Variable Bit
Rate QoS)
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-48
8/7/2019 Backbone 1
51/76
Connection-oriented data over ATM: AAL 5
SEAL the Simple and Efficient Adaptation LayerYou should realize that theres nothing inherent in AAL 3/4 that requires connectionless data service be carried. It turns out
that AAL 3/4 could carry connection-oriented data just as well. However, a number of engineers in the ATM Forum created
AAL 5, or the Simple and Efficient Adaptation Layer. It is 10% more efficient in carrying user data in cell payloads than AAL
3/4, and substantially reduces processing overhead at the receiving node.
AAL 3/4 nodes must read a sequence number, multiplexing identifier, length field, and CRC for each cell. If any cell is
received with an errored CRC, the entire PDU must be retransmitted.
Whats different about AAL 5
Theres no sequence numbers, multiplexing identifier, length, or CRC in the cell. Instead, the AAL 5 convergence layerappends a 32 bit CRC for the entire PDU, which is contained in the last cell of the segmentation and reassembly layer. This
requires that all cells of the entire convergence layer must transmitted contiguously, without other cells intervening. It also
requires that the entire convergence layer PDU be retransmitted in the event of a CRC error.
AAL 5 most common
Most ATM service providers have implemented AAL 5 in their offerings, to the exclusion of AAL 3/4. However, it appears
that AAL 3/4 will be implemented as a switching infrastructure by many SMDS service providers. They will receive SMDS
cells (Level 2 PDUs) from the CPE and encapsulate them as AAL 3/4 PDUs for transmission through the internal ATM
backbone. At the exit point from the network, the original SMDS cells will be restored, and sent to the destination CPE. Seepage 7-70 for an example.
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-49
8/7/2019 Backbone 1
52/76
Connection-oriented data over ATM: AAL 5
Information ATM Trailer8 bytes
Up to 64 Kbytes
Protocol Data UnitAAL 5 Convergence Layer
AAL 5 SAR Layer
InformationDXI Header
4 bytes
DXI Format Across
Router-DSU Interface
NH TH SH PH AH Application Data
Protocol Data UnitPayload
HeaderCell
5 bytes
HeaderCell
PayloadHeader
Cell
NotEND
NotEND END
ATM Trailer
8 bytes
Payload40 bytes + pad
CRC: 4 byt
PDU Length: 2 b
CPI: 1 byte
UUI: 1 byte
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-50
8/7/2019 Backbone 1
53/76
ATM in the Market Place
Still in infancy
Distributed Network Associates in its soon-to-be released 1998 ATM Marketing Report estimates that there are fewer than
1,000 public ATM UNI ports in service in the U.S., 16 although there are signs that activity is beginning to grow rapidly. By
comparison, the most recent study of Frame Relay ports by the same organization (January 1998) shows 350,000 in service.
Their speculation is that acceptance of ATM public network services today is roughly comparable to what we saw for Frame
Relay in 1994.
Telcos and Interexchange Carriers
Most of the regional telephone companies and the interexchange carriers in the U.S. are offering ATM service. AT&T, MCI
WorldCom, Sprint, GTE, Ameritech, Bell Atlantic, BellSouth, Pacific Bell, Southwestern Bell, SNET, and U.S. West all havesubstantial ATM infrastructure, and are aggressively marketing it to high-end business and government accounts.
Routers, DSUs, and other hardware
Most major router vendors already have available software and interface cards for ATM with either T3 or OC3 speeds. OC12
interface cards are available for the Cisco 12000 router.
Curiously, the T1 ATM is the most recent offering. Both switch vendors and carriers had been reluctant to order
With both T1 and T3 ATM access, a special ATM DSU must be used. It receives the ATM DXI frame from the router
interface, and performs the segmentation. ADC Kentrox and Digital Link have both offered T3 ADSUs (the monicker for the
ATM DSU) for three years or more. T1 ADSUs should be available before the year end.
Waiting for multimedia applications
16 UNI is the User-Network Interface, specifying how most users will connect ATM CPE to a provider network. The study did not track private
ATM growth, or NNI connections (Network-Network Interface, used between providers).
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-51
8/7/2019 Backbone 1
54/76
Most industry watchers and service providers agree that commercial ATM service wont take off until a compelling killer
applicationsuch as multimedia networkingarrives on the scene.
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-52
8/7/2019 Backbone 1
55/76
ATM in the Market Place
Less than 1,000 ATM UNI in the U.S. so says Distributed NetworkAssociates
Telcos and interexchange carriers are all offering public ATM UNI
service, from T1 to OC12
Routers are ATM-ready and DSUs are available
Waiting for multimedia applications
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-53
8/7/2019 Backbone 1
56/76
SMDSSwitched Multimegabit Data Service
What is it?
Conceived as a data networking strategy from the beginning,17 SMDS looks and feels a lot like a LAN protocol. Its
connectionless (just like Ethernet and Token Ring), supports multicasting (just like TCP/IP), and provides a variety of speeds,
ranging from 56 kbps to 34 mbps.
SMDS doesnt require individually configured PVCs to connect one router to another. Each SMDS frame (more correctly
termed a Level 3 Protocol Data Unit) contains a unique SMDS destination address for the target router. SMDS switches read
the destination address, and make packet-by-packet switching decisions (almost like routers) to forward each PDU through the
network.
Off to a slow startAlthough there are many loyal SMDS adherents, the technology has been a slow starter in the marketplace, perhaps for three
main reasons.
1. Limited geographic scope. Until 1995, SMDS networks in the U.S. could not cross local telephone companies
LATA boundaries.18
2. Expensive. Until 1994, SMDS was available only at T1 and T3 access speeds, and required pricey DSUs (which
converted SMDS packets into 53 byte cells).
3. The largest telecommunications service provider, AT&T, chose to ignore SMDS, and marketed its Frame Relay
network.
17 Unlike Frame Relay, which was conceived as an outgrowth of ISDN, and which is based on the telephony concept of circuits between two
devices exchanging information.18 MCI has offered interexchange SMDS service since 1995.
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-54
8/7/2019 Backbone 1
57/76
SMDSSwitched Multimegabit Data Service
WAN networking protocol with the look and feel of a LAN protocol
Slow start in the market
a) Until 1995, intraLATA only
b) Until 1994, T1 and T3 speeds only
c) AT&T sat out
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-55
8/7/2019 Backbone 1
58/76
Who Should Use SMDS
Whos a candidate for SMDS?Since the technology allows for connectionless (i.e.no virtual circuits) service between routers, and requires no
predetermined paths (virtual circuits) through the SMDS network, it lends itself to organizations with a very large number of
locations and to those who occasionally need to send or receive large volumes of information from outside organizations (such
as customers or vendors).
1. Full mesh networks of 20 locations or more
2. T1 speed is inadequate
3. Need to network with suppliers and customers
Full mesh is no sweatJust like Ethernet and Token Ring, SMDS easily permits any node to communicate with any other node. There is no need to
size and order individual virtual circuits. Full mesh connectivity is always provided.
High speeds
SMDS is available at 56 kbps, as well as 1.5, 4, 10, 16, 25, and 34 mbps.
Split the cost with suppliers and customers
Suppliers and customers who have SMDS access can easily internetwork with your site. Since the pricing model for SMDS
requires that each site pay only for its own service, suppliers and customers pay for their access, you pay for yours, and thereare no other charges.
1994-98 Mendes DataComm Insights
Mill Valley, California
(415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols7-56
8/7/2019 Backbone 1
59/76
Who Should Use SMDS
Full mesh, 20 sites or more
High speedfaster than T1
Connect with suppliers and customers
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-57
8/7/2019 Backbone 1
60/76
RBOC Pricing Model for SMDS
SMDS service is available in service areas of Ameritech, Bell Atlantic, BellSouth, GTE, Pacific Bell, and SNET. MCI is theonly interexchange carrier currently providing (or committed to providing) long haul service. Other interexchange carriers
seem content to wait for ATM, or to sit on the sidelines.
Pricing model for SMDS
At the local exchange level, pricing of SMDS is based on port speed into the SMDS switch plus an access channel (very much
like Frame Relay, but without the additional charge for a PVC). Access channels are defined for DS0at 56 and 64 kbps, for
DS1at 1.17 mbps, and for five classes of DS3 access (with higher prices for faster speeds).
DS3 defined classes are:
1. 4 mbps
2. 10 mbps3. 16 mbps
4. 25 mbps
5. 34 mbps
Flat rate per access line
Each SMDS access lines is priced at a flat monthly rate, with no charge for usage. Communications can be established with
any number of sites.
Approximate prices in the U.S. are $120 per month for DS0 access lines, $650 for DS1, and $2,000 to $6,000 for DS3
(depending on speed).
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-58
8/7/2019 Backbone 1
61/76
RBOC Pricing Model for SMDS
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-59
8/7/2019 Backbone 1
62/76
RBOC/GTE Availability
Available
Although SMDS is not universally available, it is available in most metropolitan areas served by four of the RBOCs, plus GTE
and SNET.
Ameritech
Bell Atlantic
Bell South
Pacific Bell
Not available
Two of the RBOCs, NYNEX and Southwestern Bell, have never offered SMDS. In addition, U.S. West announced it was
withdrawing its SMDS service in April 1996.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-60
8/7/2019 Backbone 1
63/76
RBOC/GTE Availability
Ameritech
Bell Atlantic
Bell South
GTE
Pacific Bell
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-61
8/7/2019 Backbone 1
64/76
RBOC SMDS Pricing
SMDS is priced by the RBOCs as if it were a premium DS0, DS1, or DS3 access channel. The prices presented on the facing
page include both the access line and the SMDS service with unlimited usage. Variations are due to individual RBOC pricing
strategies, and differing regulatory environments.
Flat rate
SMDS service, in the local exchange environment, is a good deal! You get an access line and a full mesh network for a flat
monthly rate for each site. Unlike the MCI interexchange service, the RBOCs do not charge on the basis of traffic volume or
usage.
Month to month vs term contract
Most RBOCs (and carriers, for that matter) offer more attractive pricing for customers willing to commit to 12 month (orlonger) contract terms.
Access line choices
Standard speeds are 56/64 kbps, 1.17 mbps (T1); and 4, 10, 16, 25, and 34 mbps (T3). Some RBOCs will provide SONET
OC3 access at 155 mbps under special terms and conditions.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-62
8/7/2019 Backbone 1
65/76
RBOC SMDS Pricing
DS0 SMDS line $120 to 225 per month per location
T1 SMDS lines $500 to 800 per month per location
T3 SMDS lines $2,000 to $6,000 (speed dependent)
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-63
8/7/2019 Backbone 1
66/76
Interexchange AvailabilityAlone among the interexchange carriers, MCI began offering SMDS nationwide in 1995. It also offers international SMDS
networkingto the U.K. (with British Telecomm), and to Ireland (with Ireland Telecomm).SMDS users can access the MCI network by either direct access (i.e.bypassing local telco SMDS switching) or through
exchange access, in which MCI connects to your local telephone company, providing only long-haul connection between sites
in different Local Access and Transport Areas (LATAs).
Direct access
The SMDS customer purchases a 56 kbps, T1, or T3 local access channel from the RBOC, connecting to the nearest MCI Point
of Presence, and selects one of the following speeds.
56/64 kbps
128 kbps
256 kbps 384 kbps
512 kbps
678 kbps
1.024 mbps
1.536 mbps
4.5 mbps
10.5 mbps
Exchange access
The local telephone company connects to MCIs SMDS network, and hands off SMDS cells bound for destinations outside the
LATA. Speeds are the same as those offered by local telcos.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-64
8/7/2019 Backbone 1
67/76
Interexchange SMDS
Direct access
a) Wide range of speeds56 kbps to 10 mbps
b) Available in 48 statesnot RBOC-dependent
Exchange accesssame speeds as RBOC/GTE access
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-65
8/7/2019 Backbone 1
68/76
MCI Pricing Model for SMDS
Pricing for both direct access and exchange access includes a fixed monthly fee for each site, plus a charge based on thevolume of data transmitted4.1 per megabyte for access speeds up to T1, and 3.0 per megabyte for 4 and 10 mbps access.
The volume charges have price caps, so that maximum montly charges are predictable.
Pricing matrix
Speed, Kbps Port Charge with
Exchange Access
Port Charge with
Direct Access
Max Volume Charge with
Exchange Access
Max Volume Charge with
Direct Access
64 $ 50 $ 180 $ 299 $ 164
128 NA 336 NA 327
256 NA 394 NA 654384 NA 578 NA 980
512 NA 735 NA 1,307
768 NA 946 NA 1,547
1,024 NA 1,178 NA 1,620
1,536 250 1,470 3,037 1,740
4,000 750 3,000 3,940 3,500
10,000 1,500 5,500 5,833 8,600
16,000 2,000 NA 8,813 NA25,000 2,500 NA 12,960 NA
34,000 3,250 NA 16,524 NA
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-66
8/7/2019 Backbone 1
69/76
MCI Pricing Model for SMDS
Port charge
a) Speed dependent
b) Access dependent
Volume charge
a) Speed dependentcap is also speed dependent, but why?
b) Access dependent
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-67
8/7/2019 Backbone 1
70/76
SMDS Design Issues
Background
Because SMDS is a connectionless service, and needs no PVCs, the network integrates much more easily with routers androuting algorithms than either Frame Relay or ATM (which are both connection-oriented networks). Routing algorithms have
no construct for an end-to-end connection through the WAN. Superimposing a connection-oriented protocol such as Frame
Relay or ATM on the inherently connectionless routing algorithm creates additional complexity and the opportunity for new
problems to arise. SMDS avoids most of the complexity and problems by mating a connectionless link layer to the
connectionless routing layer.
Each router (or a directly attached host) has a unique E.164 address (this looks like an international telephone number with
country code) assigned by the carrier or local telco. The SMDS network providers switches read the destination E.164
address in the SMDS header (see pages 68 and 69), and find the best path for delivery of each packet, in a manner not unlike
that used by routers.
Multicasting in SMDS
The ability to send multicast packetssimultaneously sending the same packet to all nodes on the networkis fully supported
in all implementations of SMDS, through the use ofgroup addresses. A number of groups can be definedone might be all
routers on the network (for OSPF link state update messages), while another group could include all intra-company locations,
and a third group, outside vendors. Group members are negotiated with the carriers implementation team during installation.
Address screening for authentication and security
Authentication and security mechanisms are implemented in the carriers switches. Each port will accept only packets with the
correct source addressauthentication. In addition, each port is configured with address screens, which only allow receptionfrom, or transmission to, a pre-determined list of source and/or destination addresses. The list is administered by the SMDS
service provider, based on the connectivity needs of each customer.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-68
8/7/2019 Backbone 1
71/76
SMDS Design Issues
Connectionless vs. connection-oriented service
Multicasting through group addresses
Address screening for authentication and security
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-69
8/7/2019 Backbone 1
72/76
SMDS Protocol
OverviewSMDS borrowed a little from here, and a little from there in order to build a specification for the technology. They borrowed
the 53 octet (byte) cell format from the Broadband ISDN and ATM development communities. And, from the IEEE 802.6
working group, they borrowed a MAC method known as distributed queue, dual busDQDB(actually invented by an
Australian firm, QPSX).
The most important concept in understanding the protocol is that, unlike PPP and Frame Relay, it is a connectionless
servicemeaning that there is no pre-established connection between pairs of sites. The SMDS switch reads each protocol
data unit (another name for a packet), and makes an instantaneous switching decision for that PDU. And it does the same for
each subsequent PDU, perhaps making several million switching decisions every second.19
How it works
The SMDS Interface Protocol is actually two protocols, defined as sub-layers within the OSI layer 2 definitions. The upper
sub-layer, called Level 3 Protocol Data Unit, can carry up to 9,188 bytes of user information; it is this packet format that
emerges from the routers serial port. The lower sub-layer is known as Level 2 Protocol Data Unit, and is responsible for
dividing the larger packets into the 53 byte cells. This part of the protocol is usually performed by a special DSUthe SMDS
DSUavailable from vendors, such as ADC Kentrox and Digital Link.
The SMDS switch receives the individual cells, reads the destination address (contained in the first few cells), and forwards
them to the next SMDS switch or directly to another user locally attached.
If one or more cells are lost, the entire Level 3 PDU must be retransmitted.
19 MCI estimates that latency encountered from New York City to Los Angeles is of the order of 40 msec.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-70
8/7/2019 Backbone 1
73/76
SMDS Protocol
InformationTrailer
38 - 50 bytes
4 bytes
Up to 9188 bytes
Level 3 Protocol Data UnitSMDS Interface Protocol
PayloadHeader T PayloadHeader T PayloadHeader TLevel 2 Protocol Data UnitSMDS Interface Protocol
InformationTrailer4 bytes
DXI Header4 bytes
DXI Format AcrossRouter-DSU Interface
NH TH SH PH AH Application Data
SMDS SMDSDA SA
SMDS SMDSDA SA
SMDS Header
AddressType
SMDS Address
4 bits 60 bits
Address type can indicate eitherindividual or group address
NOTES:
1. Cells are 53 bytes each7 bytes for the header, 44 bytes for the payload, and 2
bytes for the trailer.
2. The header of all cells from the same Level 3 PDU carry a unique message ID
number, as well as sequence numbers to enable the SMDS switch to reassemble
the Level 3 format.
1994-98 Mendes DataComm Insights
Mill Valley, California (415) 381-5500All Rights Reserved
Bridges, Switches, and Routers: Planning and Designing an Intranet
WAN Links and Backbone Protocols
7-71
8/7/2019 Backbone 1
74/76
SMDS over ATM: AAL 3/4
Conundrum of connectionless service over ATMBy definition, two end points such as hosts or routers in an ATM network are always connected by a virtual circuit,
composed of a concatenated set of virtual paths and/or virtual channels. The set may be pre-engineeredthe PVC (permanent
virtual circuit), or it may be established dynamically by the networkthe SVC (switched virtual circuit). However, a router
or host using a connectionless protocol, such as SMDS, is not prepared to request a virtual circuit when transmitting packets to
a peer; the individual packets are transmitted with an expectation of best effort delivery by the network (datagram service).
Connectionless