Ethernet Networks: Design, Implementation, Operation, Management.
Gilbert Held
Copyright  2003 John Wiley & Sons, Ltd.
ISBN: 0-470-84476-0

chapter six

Bridging and Switching
Methods and Performance
Issues
In Chapter 5, an overview of bridge operations was presented, along with
information concerning the functionality of other local area network hardware and software components. That chapter deferred until now a detailed
examination of bridging methods, to include their network use and performance issues. In this chapter, we will focus our attention on those issues,
examining different methods that bridges use for routing frames, performance
issues that govern their ability to examine and forward frames without introducing network bottlenecks, and their typical employment for interconnecting
LANs. Because LAN switches represent a special type of multiport bridge, we
will also focus our attention upon this topic later in this chapter. Thus, once
we have an appreciation for the operation and utilization of bridges, we will
turn our attention to LAN switches.

6.1 Bridging Methods
Bridges operate by examining MAC layer addresses, using the destination
and source addresses within a frame as a decision criterion to make their
forwarding decisions. Operating at the MAC layer, bridges are not addressed,
and must therefore examine all frames that flow on a network. Because bridges
operate at the MAC layer, they in effect terminate a collision domain. That is,
if a collision is detected upon one port of a bridge, it is not propagated onto
any output port. This means that, unlike a repeater, a bridge can be used to
extend the span of a LAN.

279


Ethernet 3

Ethernet 2

Ethernet 1

E

D

F

B

Port 1

Bridge 1

Bridge 1 port / address table
Port 1
A
B

Port 2
C
D
E
F


are shown and labeled on each local area network. Those labels represent the
48-bit MAC address of each station.

Port/Address Table Construction
As previously noted in Chapter 5, a bridge constructs its port/address table
by using what can be referred to as the ‘‘three F’s’’ — flooding, filtering, and
forwarding. If a bridge encounters a frame with a destination address that is
not in its port/address table, it transmits the frame onto all other ports except
the port it was received on. If the destination address is in its port/address
table and does not represent the port the frame was received on, the bridge
forwards the frame onto the port corresponding to the entry in the table for
the destination address. If the destination address is in the port/address table
and represents the port the frame was received on, there is no need to forward
the frame. Thus, the frame is filtered by the bridge.
In examining the construction of bridge port/address tables for the network
shown in Figure 6.1, we will assume that each bridge operates as a transparent
bridge. As frames flow on the Ethernet, bridge 1 examines the source address
of each frame. Eventually, after both stations A and B have become active, the
bridge associates their address as being on port 1 of that device. Any frames
with a destination address other than stations A or B are considered to be on
another network. Thus, bridge 1 would eventually associate addresses C, D,
E, and F with port 2, once it receives frames with those addresses in their
destination address fields. Similarly, bridge 2 constructs its own port/address
table. Since frames from Ethernet 1 and 1 Ethernet 2 can have source addresses
of A, B, C, or D, eventually the port/address table of bridge 2 associates those
addresses with port 1 of that device. Since frames from Ethernet 1 or Ethernet 2
with a destination address of E or F are not on those local area networks,
bridge 2 then associates those addresses with port 2 of that device.
The port/address tables previously shown in Figure 6.1 are normally stored
in bridge memory sorted by MAC address. In addition, the time the entry

Ethernet 2
B

A

E

Port 1

Bridge 1

F

Port 2

Port 1

Port 1

Bridge 2

Bridge 3

Port 2

Port 2

Ethernet 3
C


the 802.1D standard with an intelligent algorithm known as the Spanning
Tree Protocol (STP). The STP, based on graph theory, converts a loop into a
tree topology by disabling a link. This action ensures there is a unique path
from any node in an intranet to every other node. Disabled nodes are then
kept in a standby mode of operation until a network failure occurs. At that
time, the STP will attempt to construct a new tree using any of the previously
disabled links.

Operation
To illustrate the operation of the STP, we must first become familiar with
the difference between the physical and active topology of bridged networks.
In addition, there are a number of terms associated with the spanning tree
algorithm, as defined by the protocol, that we should become familiar with.
Thus, we will also review those terms before discussing the operation of
the algorithm.

Physical versus Active Topology
In transparent bridging, a distinction is made between the physical and active
topology resulting from bridged local area networks. This distinction enables
the construction of a network topology in which inactive but physically


284

chapter six

constructed routes can be placed into operation if a primary route should fail,
and in which the inactive and active routes would form an illegal circular
path violating the spanning tree algorithm if both routes were active at the
same time.

the weight of a tree represents the sum of its links or edges. If the weight
or length of the links or tree edges differ, then different tree structures will
have different weights. Thus, the identification of the minimum spanning tree
requires us to examine each of the spanning trees supported by a graph and
identify the structure that has the minimum length or weight.


bridging and switching methods and performance issues

285

Token-ring 1

B1
C =10

Ethernet 2

Ethernet 1

B2
C =10

B4
C =15

B3
C =15

Token-ring 2

Figure 6.3 Physical versus active topology. When transparent bridges are
used, the active topology cannot form a closed loop in the intranet.


286

chapter six

(a) Network graph

(b) Possible spanning trees

Figure 6.4

Forming spanning trees from a network graph.

The identification of the minimum spanning tree can be accomplished
by listing all spanning trees and finding the minimum weight or length
associated with the list. This is a brute force method that always works but
is not exactly efficient, especially when a graph becomes complex and can
contain a significant number of trees. A far better method is obtained by the
use of an appropriate algorithm.

Kruskal’s Algorithm
There are several popular algorithms developed for solving the minimum
spanning tree of a graph. One of those algorithms is the Kruskal algorithm
which is relatively easy to understand and will be used to illustrate the
computation of a minimum spanning tree. Because we need weights or
lengths assigned to each edge or link in a graph, let’s revise the network graph
previously shown in Figure 6.4 and add some weights. Figure 6.5 illustrates

1. Sort the edges of the graph (G) in their increasing order by weight
or length.
2. Construct a subgraph (S) of G and initially set it to the empty state.
3. For each edge (e) in sorted order:
If the endpoints of the edges (e) are disconnected in S, add them to S.
Using the graph shown in Figure 6.5, let’s apply Kruskal’s algorithm
as follows:
1. The sorted edges of the graph in their increasing order by weight or
length produces the following table:
Edge

Weight/Length

A-C

1

B-D

3

C-B

4

C-D

6

A-B

or

The third operation produces:

S = (A,B) + (B,D) + (C,B)

A

B

C

D

or

Note that we cannot continue as the endpoints in S are now all connected.
Thus, the minimum spanning tree consists of the edges or links (A, B) +
(B, D) + (C, B) and has the weight 1 + 4 + 3, or 7. Now that we have an
appreciation for the method by which a minimum spanning tree is formed, let’s
turn our attention to its applicability in transparent bridge-based networks.
Similar to the root of a tree, one bridge in a spanning tree network will
be assigned to a unique position in the network. Known as the root bridge,
this bridge is assigned as the top of the spanning tree, and because of this
position, it has the potential to carry the largest amount of intranet traffic due
to its position.
Because bridges and bridge ports can be active or inactive, a mechanism
is required to identify bridges and bridge ports. Each bridge in a spanning
tree network is assigned a unique bridge identifier. This identifier is the MAC
address on the bridge’s lowest port number and a two-byte bridge priority

bridges linking two networks will not forward frames and will be in a blocking
state of operation.

Constructing the Spanning Tree
The spanning tree algorithm employs a three-step process to develop an active
topology. First, the root bridge is identified. To accomplish this, each bridge
in the intranet will initially assume it is the root bridge. To determine which
bridge should actually act as the root bridge, each bridge will periodically
transmit bridge protocol data unit (BPDU) frames that are described in the
following section. BPDU frames under Ethernet version 2 are referred to as
HELLO frames or messages and are transmitted on all bridge ports. Each
BPDU frame includes the priority of the bridge defined at installation time. As
the bridges in the intranet periodically transmit their BPDU frames, bridges
receiving a BPDU with a lower priority value than its own cease transmitting
their BPDUs; however, they forward BPDUs with a lower priority value.
Thus, after a short period of time the bridge with the lowest priority value
is recognized as the root bridge. In Figure 6.3b we will assume bridge 1 was
selected as the root bridge. Next, the path cost from each bridge to the root
bridge is determined, and the minimum cost from each bridge becomes the
root path cost. The port in the direction of the least path cost to the root
bridge, known as the root port, is then determined for each bridge. If the root
path cost is the same for two or more bridges linking LANs, then the bridge
with the highest priority will be selected to furnish the minimum path cost.
Once the paths are selected, the designated ports are activated.


290

chapter six


configuration BPDUs.
When a standby bridge is required to assume the role of the root or designated
bridge, the HELLO BPDU will indicate that a standby bridge should become
a designated bridge. The process by which bridges determine their role in
a spanning tree network is iterative. As new bridges enter a network, they
assume a listening state to determine their role in the network. Similarly,
when a bridge is removed, another iterative process occurs to reconfigure the
remaining bridges.


bridging and switching methods and performance issues

291

Although the STP algorithm procedure eliminates duplicate frames and
degraded intranet performance, it can be a hindrance for situations where
multiple active paths between networks are desired. In addition, if a link or
device fails, the time required for a new spanning tree to be formed via the
transmission of BPDUs can easily require 45 to 60 seconds or more. Another
disadvantage of STP occurs when it is used in remote bridges connecting
geographically dispersed networks. For example, returning to Figure 6.2,
suppose Ethernet 1 were located in Los Angeles, Ethernet 2 in New York, and
Ethernet 3 in Atlanta. If the link between Los Angeles and New York were
placed in a standby mode of operation, all frames from Ethernet 2 routed to
Ethernet 1 would be routed through Atlanta. Depending on the traffic between
networks, this situation might require an upgrade in the bandwidth of the links
connecting each network to accommodate the extra traffic flowing through
Atlanta. Since the yearly cost of upgrading a 56- or 64-Kbps circuit to a 128Kbps fractional T1 link can easily exceed the cost of a bridge or router, you
might wish to consider the use of routers to accommodate this networking
situation. In comparison, when using local bridges, the higher operating

phase IV talk
Ethernet NIC
Ethernet

B
IEEE 802.3
M

M
Frame differences
Ethernet
Type

Information

IEEE 802.3
Length

Information

DSAP SSAP Control

Information

Legend:
= Workstation
M = Macintosh

Figure 6.6 Protocol differences preclude linking IEEE 802.3 and Ethernet
networks using transparent bridges. A Macintosh computer connected on an

later in this section.

Field format
Up to 16 bytes

2 bytes
Control

Ring #

Bridge #

B B B L L L L L D LF LF LF LF R R R

Ring #

Ring #
12 bits

Bridge #

Bridge #
4 bits

B are broadcast bits
Bit settings
Designator
Nonbroadcast
0XX
All-routes broadcast

294

chapter six

The control field contains information that defines how information will be
transferred and interpreted and what size the remainder of the RIF will be. The
three broadcast bit positions indicate a nonbroadcast, all-routes broadcast, or
single-route broadcast situation. A nonbroadcast designator indicates a local
or specific route frame. An all-routes broadcast designator indicates that a
frame will be transmitted along every route to the destination station. A
single-route broadcast designator is used only by designated bridges to relay
a frame from one network to another. In examining the broadcast bit settings
shown in Figure 6.7, note that the letter X indicates an unspecified bit setting
that can be either a 1 or 0.
The length bits identify the length of the RIF in bytes, while the D bit
indicates how the field is scanned, left to right or right to left. Since vendors
have incorporated different memory in bridges which may limit frame sizes,
the LF bits enable different devices to negotiate the size of the frame. Normally,
a default setting indicates a frame size of 512 bytes. Each bridge can select
a number, and if it is supported by other bridges, that number is then used
to represent the negotiated frame size. Otherwise, a smaller number used
to represent a smaller frame size is selected, and the negotiation process is
repeated. Note that a 1500-byte frame is the largest frame size supported
by Ethernet IEEE 802.3 networks. Thus, a bridge used to connect Ethernet
and Token-Ring networks cannot support the use of Token-Ring frames
exceeding 1500 bytes.
Up to eight route number subfields, each consisting of a 12-bit ring number
and a 4-bit bridge number, can be contained in the routing information field.
This permits two to eight route designators, enabling frames to traverse up
to eight rings across seven bridges in a given direction. Both ring numbers


B1

D

New York
B3

Atlanta

B5
0 R3 B5 R2 B4 R2 B3 R1 A

B3 R1 A

R2

B4

B4 R2 B3 R1 A

Figure 6.8 Source routing discovery operation. The route discovery process
results in each bridge entering the originating ring number and its bridge
number into the RIF.

When a station wants to originate communications, it is responsible for
finding the destination by transmitting a discovery packet to network bridges
and other network stations whenever it has a message to transmit to a new
destination address. If station A wishes to transmit to station C, it sends a
route discovery packet containing an empty RIF and its source address, as

There are several advantages associated with source routing. One advantage is
the ability to construct mesh networks with loops for a fault-tolerant design;
this cannot be accomplished with the use of transparent bridges. Another
advantage is the inclusion of routing information in the information frames.
Several vendors have developed network management software products that
use that information to provide statistical information concerning intranet
activity. Those products may assist you in determining how heavily your
wide area network links are being used, and whether you need to modify the
capacity of those links; they may also inform you if one or more workstations
are hogging communications between networks.

Disadvantages
Although the preceding advantages are considerable, they are not without
a price. That price includes a requirement to identify bridges and links
specifically, higher bursts of network activity, and an incompatibility between
Token-Ring and Ethernet networks. In addition, because the structure of the
Token-Ring RIF supports a maximum of seven entries, routing of frames is
restricted to crossing a maximum of seven bridges.
When using source routing bridges to connect Token-Ring networks, you
must configure each bridge with a unique bridge/ring number. In addition,
unless you wish to accept the default method by which stations select a frame
during the route discovery process, you will have to reconfigure your LAN
software. Thus, source routing creates an administrative burden not incurred
by transparent bridges.
Due to the route discovery process, the flooding of discovery frames occurs in
bursts when stations are turned on or after a power outage. Depending upon
the complexity of an intranet, the discovery process can degrade network


bridging and switching methods and performance issues

and processes the packet using a transparent mode of operation, using routing
tables generated by a spanning tree algorithm.

Advantages
There are several advantages associated with source routing transparent
bridges. First and perhaps foremost, they enable different networks to use
different local area network operating systems and protocols. This capability enables you to interconnect networks developed independently of one
another, and allows organization departments and branches to use LAN
operating systems without restriction. Secondly, also a very important consideration, source routing transparent bridges can connect Ethernet and


298

chapter six

Token-Ring networks while preserving the ability to mesh or loop TokenRing networks. Thus, their use provides an additional level of flexibility for
network construction.

Translating Operations
When interconnecting Ethernet/IEEE 802.3 and Token-Ring networks, the
difference between frame formats requires the conversion of frames. A bridge
that performs this conversion is referred to as a translating bridge.
As previously noted in Chapter 4, there are several types of Ethernet frames,
such as Ethernet, IEEE 802.3, Novell’s Ethernet-802.3, and Ethernet-SNAP.
The latter two frames represent variations of the physical IEEE 802.3 frame
format. Ethernet and Ethernet-802.3 do not use logical link control, while IEEE
802.3 CSMA/CD LANs specify the use of IEEE 802.2 logical link control. In
comparison, all IEEE 802.5 Token-Ring networks either directly or indirectly
use the IEEE 802.2 specification for logical link control.
The conversion from IEEE 802.3 to IEEE 802.5 can be accomplished by

Legend:
DA
SA
AC
FC
RIF
DSAP
SSAP
ED
FS

= Destination Address
= Source Address
= Access Control
= Frame Control
= Routing Information Field
= Destination Service Access Point
= Source Service Access Point
= End Delimiter
= Frame Status Field

Figure 6.9

IEEE 802.3 to 802.5 frame conversion.


bridging and switching methods and performance issues

299


Legend:
DA
SA
AC
FC
RIF
DSAP
SSAP
ED
FS
OC

Figure 6.10

= Destination Address
= Source Address
= Access Control
= Frame Control
= Routing Information Field
= Destination Service Access Point
= Source Service Access Point
= End Delimiter
= Frame Status Field
= Organization Code

Ethernet to Token-Ring frame conversion.

linking an IEEE 802.3 network to an IEEE 802.5 network. Note that fields
unique to the IEEE 802.3 frame are discarded, while fields common to both
frames are copied. Fields unique to the IEEE 802.5 frame are inserted by

in the middle portion of Figure 6.11. This extension results in sequential or
cascaded bridging, and is appropriate when intra-LAN traffic is necessary but
minimal. This intranet topology is also extremely useful when the length of an
Ethernet must be extended beyond the physical cabling of a single network. By
locating servers appropriately within each network segment, you may be able
to minimize inter-LAN transmission. For example, the first network segment
could be used to connect marketing personnel, while the second and third
segments could be used to connect engineering and personnel departments.
This might minimize the use of a server on one network by persons connected
to another network segment.
A word of caution is in order concerning the use of bridges. Bridging
forms what is referred to as a flat network topology, because it makes its
forwarding decisions using layer 2 MAC addresses, which cannot distinguish
one network from another. This means that broadcast traffic generated on
one segment will be bridged onto other segments which, depending upon
the amount of broadcast traffic, can adversely affect the performance on
other segments.
The only way to reduce broadcast traffic between segments is to use a
filtering feature included with some bridges or install routers to link segments. Concerning the latter, routers operate at the network layer and forward
packets explicitly addressed to a different network. Through the use of network addresses for forwarding decisions, routers form hierarchical structured
networks, eliminating the so-called broadcast storm effect that occurs when
broadcast traffic generated from different types of servers on different segments
are automatically forwarded by bridges onto other segments.
Both serial and sequential bridging are applicable to transparent, source
routing, and source routing transparent bridges that do not provide redundancy nor the ability to balance traffic flowing between networks. Each of these
deficiencies can be alleviated through the use of parallel bridging. However,


bridging and switching methods and performance issues


The lower portion of Figure 6.11 illustrates the use of parallel bridges to
interconnect two Token-Ring networks. This bridging configuration permits


302

chapter six

one bridge to back up the other, providing a level of redundancy for linking
the two networks as well as a significant increase in the availability of one
network to communicate with another. For example, assume the availability
of each bridge used at the top of Figure 6.11 (serial bridging) and bottom
of Figure 6.11 (parallel bridging) is 90 percent. The availability through two
serially connected bridges would be 0.9 × 0.9 (availability of bridge 1 ×
availability of bridge 2), or 81 percent. In comparison, the availability through
parallel bridges would be 1 − (0.1 × 0.1), which is 99 percent.
The dual paths between networks also improve inter-LAN communications
performance, because communications between stations on each network can
be load balanced. The use of parallel bridges can thus be expected to provide a
higher level of inter-LAN communications than the use of serial or sequential
bridges. However, as previously noted, this topology is not supported by
transparent bridging.

Star Bridging
With a multiport bridge, you can connect three or more networks to form a
star intranet topology. The top portion of Figure 6.12 shows the use of one
bridge to form a star topology by interconnecting four separate networks.
This topology, or a variation on this topology, could be used to interconnect
networks on separate floors within a building. For example, the top network
could be on floor N + 1, while the bottom network could be on floor N − 1 in


B

Legend:
B

= Workstation
= Bridge

Figure 6.12

Star and backbone bridging.

6.3 Bridge Performance Issues
The key to obtaining an appropriate level of performance when interconnecting networks is planning. The actual planning process will depend upon
several factors, such as whether separate networks are in operation, the type
of networks to be connected, and the type of bridges to be used — local
or remote.

Traffic Flow
If separate networks are in operation and you have appropriate monitoring
equipment, you can determine the traffic flow on each of the networks to be