Charles M. Kozierok The TCP-IP Guide

Подождите немного. Документ загружается.

Furthermore, it is normal for the delay on any link to vary over time; in the case where there

are two routes that are similar in cost, fluctuations in the delay for each route could result in

rapid changes between routes (a phenomenon sometimes called route flapping). Adjust-

ments are needed to the basic overview of the operation of the HELLO protocol above, to

avoid these sorts of problems.

Current Role in TCP/IP

Like other early routing protocols, HELLO includes nothing fancy like authentication and so

on; these features were not really needed in the early days of the Internet, when the inter-

networks were small and easily controlled. As the Internet grew, HELLO was eventually

replaced by newer routing protocols such as RIP. It is now considered a historical protocol

(read: obsolete) and is no longer used.

Key Concept: The HELLO Protocol was used on very early routers on the

precursors of the Internet to exchange routing information. It is a distance-vector

protocol like RIP and GGP but differs in that it uses calculated delay as a metric

instead of hop count. Like GGP, it is now considered a historical protocol and is no longer

part of TCP/IP.

Interior Gateway Routing Protocol (IGRP)

If you have read a substantial portion of this Guide already, you have probably noticed that I

greatly prefer universal, open standards to proprietary standards. (I explain the reasons

why in the Networking Fundamentals section on standards.) I am far from alone in this view,

and in fact, it's no exaggeration to say that much of the success of TCP/IP and the Internet

is tied to the fact that they were both developed using the open RFC process, and still are to

this day.

That said, there are certain situations where a proprietary protocol can be of benefit, and

can even achieve considerable success, if a minimum of two factors are true. First, there

must be either a lack of a suitable open protocol, or a gap in the feature coverage of

existing open protocols that creates an “opportunity” for a proprietary protocol to succeed.

Second, the proprietary protocol must be either initiated or strongly supported by a “big

player” in the industry, to help ensure that other companies will take notice and give the

protocol a chance to become a standard.

This situation arose in the 1980s in the world of routing protocols. At that time, the most

popular interior routing protocol was the Routing Information Protocol (RIP). As described in

the RIP section of this Guide, RIP does a basically good job, but has a number of limitations

and problems that are inherent to the protocol and not easily resolved. In the mid-1980s,

open alternatives like OSPF did not yet exist; even if it had, OSPF is much more complex

than RIP and therefore sometimes not a good alternative.

Cisco Systems, definitely one of the “big names” in networking and especially internet-

working and routing, decided to develop a new routing protocol that would be similar to RIP

but would provide greater functionality and solve some of RIP's inherent problems: the

Interior Gateway Routing Protocol (IGRP). IGRP—which conveniently uses both the words

“gateway” and “routing” in its name to convey the equivalence of the two words in internet-

working standards—was designed specifically to be a replacement for RIP. It is similar in

many ways, and keeps RIP's simplicity, one of its key strengths. At the same time, IGRP

overcomes two key limitations of RIP: the use of only hop count as a routing metric, and the

hop count limit of 15.

Overview of Operation

Like RIP, IGRP is a distance-vector routing protocol designed for use with an autonomous

system, and thus uses the same basic mechanism for route determination. Each router

routinely sends out on each local network to which it is attached a message containing a

copy of its routing table. This message contains pairs of reachable networks and costs

(metrics) to reach each network. A router receiving this message knows it can reach all the

networks in the message as long as it can reach the router that sent the message. It

computes the cost to reach those networks by adding to their costs, the cost to reach the

router that sent the message. The routers update their tables accordingly, and send this

information out in their next routine update. Eventually, each router in the autonomous

system has information about the cost to reach each network in it.

Features and Capabilities

An important difference between RIP and IGRP, however, is that where RIP only allows the

cost to reach a network to be expressed in terms of hop count, IGRP provides a much more

sophisticated metric. In IGRP, the overall cost to reach a network is computed based on

several individual metrics, including internetwork delay, bandwidth, reliability and load. The

calculation of cost can be customized by an administrator, who can set relative weightings

to the component metrics to reflect the priorities of that autonomous system. So, if a

particular administrator feels route cost would be best minimized by emphasizing reliability

over bandwidth, he or she can do this. Such a system provides tremendous flexibility over

the rigid hop-count system of RIP. Unlike RIP, IGRP also does not have any inherent limit of

15 hops between networks.

To this basic algorithm, IGRP adds a feature called multipath routing. This allows multiple

paths between routes to be used automatically, with traffic shared between them. The traffic

can either be shared evenly, or apportioned unevenly based on the relative cost metric of

each path. This provides improved performance and again, flexibility.

Since IGRP is a distance-vector protocol like RIP, it shares many of RIP's algorithmic

“issues”. Unsurprisingly, then, IGRP must incorporate many of the same stability features

as RIP, including the use of split horizon, split horizon with poisoned reverse (in certain

circumstances) and the employment of hold-down timers. Like RIP, IGRP also uses timers

to control how often updates are sent, how long routers are “held down”, and how long

routes are held in the routing table before being expired.

Cisco originally developed IGRP for IP networks, and since IP is predominant in the

industry, this is where it is most often seen. IGRP is not specific to IP, however, and can be

used with other internetworking protocols if implemented for them. As we will see in the next

topic, Cisco also used IGRP as the basis for an improved routing protocol called EIGRP,

developed several years after the original.

Key Concept: In the 1980s, Cisco Systems created the Interior Gateway Routing

Protocol (IGRP) as an improvement over the industry standard protocol RIP. IGRP is

a distance-vector protocol like RIP and similar to it in many ways, but includes

several enhancements. Amongst the most important of these is an elimination of the 15 hop

limit between routers, and the ability to use metrics other than hop count to determine

optimal routes.

Enhanced Interior Gateway Routing Protocol (EIGRP)

Cisco Systems, a leader in the world of internetworking and routing technology, developed

the Interior Gateway Routing Protocol (IGRP) in the mid-1980s as an alternative protocol to

RIP for use in TCP/IP autonomous systems. IGRP represented a substantial improvement

over RIP, but like any successful company, Cisco was not content to rest on its laurels.

Cisco knew that IGRP had significant room for improvement, so they set to work on creating

a better version of IGMP in the early 1990s. The result was the Enhanced Interior Gateway

Routing Protocol (EIGRP).

Comparing IGRP and EIGRP

Compared to the original protocol, EIGRP is more of an evolution than a revolution. EIGRP

is still a distance-vector protocol, but is more sophisticated than other distance-vector

protocols like IGRP or RIP, and includes certain features that are more often associated

with link-state routing protocols like OSPF than distance-vector algorithms. Also, since

Cisco realized many of the organizations deciding to use EIGRP would be migrating to it

from IGRP, they took special steps to maximize compatibility between the two.

The chief differences between IGRP and EIGRP are not in what they do, but how they do it.

EIGRP changes the way that routes are calculated, in an effort to improve efficiency and

the speed of route convergence (agreement on routes between different routers in the inter-

network). EIGRP is based on a new route calculation algorithm called the Diffusing Update

Algorithm (DUAL), developed at a company called SRI International by Dr. J. J. Garcia-

Luna-Aceves.

EIGRP’s DUAL Route Calculation Algorithm

DUAL differs from a typical distance-vector algorithm primarily in that it maintains more

topology information about the internetwork than that used by protocols like RIP or IGRP. It

uses this information to automatically select least-cost, loop-free routes between networks.

EIGRP uses a metric that combines an assessment of the bandwidth of a link and the total

delay to send over the link. (Other metrics are configurable as well, though not recom-

mended.) When a neighboring router sends changed metric information, routes are

recalculated and updates sent as needed. DUAL will query neighboring routers for reach-

ability information if needed (for example, if an existing route fails).

This “as needed” aspect of operation highlights an important way that EIGRP improves

performance over IGRP. EIGRP does not send routine route updates, only partial updates

as required, reducing the amount of traffic generated between routers. Furthermore, these

updates are designed so that only the routers that need the updated information receive

them.

Other Features of EIGRP

In order to build the tables of information it needs to calculate routes, EIGRP requires that

each router make and maintain contact with other routers on their local networks. To facil-

itate this, EIGRP incorporates a neighbor discovery/recovery process. This system involves

the exchange of small Hello messages that let routers discover which other routers on the

local network, and to periodically check their reachability. This is very similar to the use of

the identically-named Hello messages in OSPF, and has a low impact on bandwidth use

because the messages are small and infrequently sent.

Some of the features in IGRP carry through to its successor, such as the use of split horizon

with poisoned reverse for improved stability. In addition to the basic improvements of

efficiency and route convergence that accrue from the algorithm itself, EIGRP includes

some other features. These include support for variable length subnet masks (VLSM) and

support for multiple network-layer protocols. This means that EIGRP could be configured to

function on a network running IP as well as another layer three protocol.

Key Concept: Developed in the 1990s, the Enhanced Interior Gateway Routing

Protocol (EIGRP) is an improved version of Cisco’s IGRP. It is similar to IGRP in

many respects, but uses a more sophisticated route calculation method called the

Diffusing Update Algorithm (DUAL). EIGRP also includes several features that make it

more intelligent in how it computes routes, borrowing concepts from link-state routing

protocols, and using more efficient partial updates rather than sending out entire routing

tables.

TCP/IP Exterior Gateway/Routing Protocols (BGP and EGP)

For ease of administration, routers on large internetworks are grouped into autonomous

systems (ASes) that are independently controlled by different organizations and

companies. Interior routing protocols such as RIP and OSPF are used to communicate

routing information between routers within an autonomous system. Obviously, if interior

routing protocols are used within ASes, we need another set of routing protocols to send

that information between ASes. These are called exterior routing protocols.

The entire point of autonomous system architecture is conveyed in the meaning of the first

word in that phrase: autonomous. The details of what happens within an AS are hidden

from other ASes, which allows the administrator of an AS to have the independence to

control how he or she runs it, including the selection of one or more from a variety of

different interior routing protocols. In contrast, to reliably connect ASes together, it is

essential that each one be running the same exterior routing protocol, or the result would be

something akin to the Tower of Babel. The result of this is that in TCP/IP there is generally

only one exterior routing protocol in widespread use at a given time.

In this section I describe two different TCP/IP exterior routing protocols. The first is the

Border Gateway Protocol (BGP), the one used in modern TCP/IP. BGP is very important

since it is used on the current Internet and other larger internetworks, so it is covered in

considerable detail. The second is the Exterior Gateway Protocol (EGP). This is an

obsolete protocol that was used for communication between non-core routers and the

router core in the early Internet, and is described briefly for both completeness and

historical interest.

Background Information: I am assuming in this section that you are already at

least somewhat familiar with interior routing protocols, at least to the extent of

understanding what they do in basic terms. If you have not yet read up on RIP, the

most common interior routing protocol, you may wish to skim that section. At the very least,

make sure you are familiar with the overview of routing protocol architectures.

TCP/IP Border Gateway Protocol (BGP/BGP-4)

Modern TCP/IP internetworks are comprised of autonomous systems (ASes) that are run

independently. Each may use an interior routing protocol such as RIP, OSPF, IGRP or

EIGRP to select routes between networks within the AS. To form larger internetworks, and

especially the “mother of all internetworks”, the Internet, these autonomous systems must

be connected together. This requires use of a consistent exterior routing protocol that all

ASes can agree upon, and in today's TCP/IP that protocol is the Border Gateway Protocol

(BGP).

In this section I describe the characteristics, general operation and detailed operation of the

Border Gateway Protocol (BGP). The discussion is divided into two subsections. The first

provides an overview and general look at the operation of BGP, including a discussion of

key concepts such as topology, neighbor relationships, route determination and general

messaging. The second gives a more detailed analysis of the different message types and

how they are used, and describes the format of each message as well.

BGP is another in the rather large group of protocols and technologies that is so complex it

would take dozens of topics to do justice. Therefore, I include here my somewhat standard

disclaimer that you will find in this section only a relatively high-level look at BGP. You will

need to refer to the BGP standards (described in the topic on BGP standards and versions)

if you need more details.

Note: The current version of BGP is version 4, also called BGP-4. This is the only

version widely used today, so unless otherwise indicated, assume BGP-4

wherever you see “BGP”.

BGP Fundamentals and General Operation

If you were to ask the average Internet user, or even that typical network administrator, to

make a list of the ten most important TCP/IP protocols, it's probable that BGP wouldn't often

show up. Routing protocols are “worker bees” of the TCP/IP suite and just not very exciting.

The reality, however, is that BGP is a critically important protocol to the operation of larger

internetworks and the Internet itself. It is the “glue” that binds smaller internetworks (auton-

omous systems) together, and it ensures that every organization is able to share routing

information. It is this function that lets us take disparate networks and internetworks and find

efficient routes from any host to any other host, regardless of location.

In this section I begin our look at the very important BGP protocol by covering its basic

concepts and general operation. I start, as usual, with an overview of the protocol and

discussion of its history, standards and versions, including a discussion of its key features

and characteristics. I then cover basic operational concepts, including topology, the notion

of BGP speakers and neighbor relationships. I discuss BGP traffic types and how policies

can be used to control traffic flows on the internetwork. I explain how BGP routers store and

advertise routes, and the function of Routing Information Bases (RIBs). I describe the basic

algorithm used by BGP and how path attributes describe routes. I then provide a summary

of how the BGP route selection process operates. I conclude with a general description of

BGP's operation and its high-level use of various messages.

BGP Overview, History, Standards and Versions

As I describe briefly in the overview of routing protocol concepts, the way that routers were

connected in the early Internet was quite different than it is today. The early Internet had a

set of centralized routers functioning like a “core” autonomous system. These routers used

the Gateway-to-Gateway Protocol for communication between them within the AS, and the

aptly-named Exterior Gateway Protocol (EGP) to talk to routers outside the core.

Motivation and Development of BGP

When the Internet grew and moved to autonomous system (AS) architecture, EGP was still

able to function as the exterior routing protocol for the Internet. However, as the number of

autonomous systems in an internetwork grows, the importance of communication between

them grows as well. EGP was functional but had several weaknesses that became more

problematic as the Internet grew in size. It was necessary to define a new exterior routing

protocol that would provide enhanced capabilities for use on the growing Internet.

In June 1989, the first version of this new routing protocol was formalized, with the

publishing of RFC 1105, A Border Gateway Protocol (BGP)

. This initial version of the BGP

standard defined most of the concepts behind the protocol, as well as key fundamentals

such as messaging, message formats and how devices operate in general terms. It estab-

lished BGP as the Internet's exterior routing protocol of the future.

BGP Evolution, Versions and Defining Standards

Due to the importance of a protocol that spans the Internet, work continued on BGP for

many years after the initial standard was published. The developers of BGP had to correct

problems with the initial protocol, refine BGP's operation, improve efficiency, and add

features. It was also necessary to make adjustments to allow BGP to keep pace with other

changes in the TCP/IP protocol suite, such as the invention of classless addressing and

routing.

The result of this ongoing work is that BGP has evolved through several versions and

standards. These are sometimes called BGP-N where N is the version number. Table 134

shows the history of BGP standards, providing the RFC numbers and names and a brief

summary of the changes made in each version.

Table 134: Border Gateway Protocol (BGP) Versions and Defining Standards

RFC

Number

Date Name

BGP

Version

Description

1105

June

1989

A Border Gateway

Protocol (BGP)

BGP-1 Initial definition of the BGP protocol.

1163

June

1990

A Border Gateway

Protocol (BGP)

BGP-2

This version cleaned up several issues with BGP-1,

and refined the meaning and use of several of the

message types. It also added the important concept

of path attributes, which communicate information

about routes.

BGP-1 was designed around the notion of a direc-

tional topology, with certain routers being “up”,

“down” or “horizontal” relative to each other; BGP-2

removed this concept, making BGP better suited to

an arbitrary AS topology.

Note that the RFC title is not a typo; they didn't put

“version 2” in the title.

1267

October

1991

Border Gateway

Protocol 3 (BGP-3)

BGP-3

This version optimized and simplified route infor-

mation exchange, adding an identification capability

to the messages used to establish BGP communi-

cations, and incorporating several other

improvements and corrections.

(They left the “A” off the title of this one for some

reason. ☺)

1654

July

1994

A Border Gateway

Protocol 4 (BGP-4)

BGP-4

Initial standard for BGP-4, revised in RFC 1771.

See just below.

1771

March

1995

A Border Gateway

Protocol 4 (BGP-4)

BGP-4

Current standard for BGP-4. The primary change in

BGP-4 is support for Classless Inter-Domain

Routing (CIDR). The protocol was changed to allow

prefixes to be specified that represent a set of

aggregated networks. Other minor improvements

were also made to the protocol.

As you might imagine, changing the version of a protocol like BGP is not an easy under-

taking. Any modification of the protocol would require the coordination of many different

organizations. The larger the Internet grows, the more difficult this would be. As a result,

despite frequent version changes in the early 1990s, BGP-4 remains today the current

version of the standard, and is the one that is widely used. Unless otherwise specified, any

mention of BGP in this Guide refers to BGP-4.

BGP Supplementary Standards

Supplementing RFC 1771 are three other consecutively-numbered RFCs published simul-

taneously with it, which provide supporting information about BGP's functions and use, as

shown in Table 135.

Overview of BGP Functions and Features

If I were to summarize the job of BGP in one phrase, it would be “to exchange network

reachability information between autonomous systems and from this information determine

routes to networks”. In a typical internetwork (and in the Internet) each autonomous system

designates one or more routers that run BGP software. BGP routers in each autonomous

system are linked to those in one or more other autonomous systems. Each BGP stores

information about networks and the routes to them in a set of Routing Information Bases

(RIBs). This route information is exchanged between BGP routers, and propagated

throughout the entire internetwork, allowing each AS to find paths to each other AS, and

thereby enabling routing across the entire internetwork.

Table 135: Additional Defining Standards For BGP-4

RFC

Number

Name Description

1772

Application of the Border

Gateway Protocol in the

Internet

Provides additional conceptual information on the

operation of BGP and how it is applied to and used on the

Internet.

This document is sometimes considered a companion of

RFC 1771 with the pair defining BGP-4.

1773

Experience with the BGP-4

Protocol

Describes the experiences of those testing and using

BGP-4, and provides information that justified its accep-

tance as a standard.

1774 BGP-4 Protocol Analysis

Provides more detailed technical information about the

operation of the BGP-4 protocol.

Key Concept: The exterior routing protocol used in modern TCP/IP internetworks is

the Border Gateway Protocol (BGP). Initially developed in the late 1980s as a

successor to EGP, BGP has been revised many times; the current version is 4, so

BGP is also commonly called BGP-4. BGP’s primary function is the exchange of network

reachability information between autonomous systems to allow each AS on an internetwork

to send messages efficiently to every other one.

BGP supports an arbitrary topology of ASes, meaning they can be connected in any

manner. An AS must have a minimum of one router running BGP, but can have more than

one. It is also possible to use BGP to communicate between BGP routers within the same

autonomous system.

BGP uses a fairly complex system for route determination. The protocol goes beyond the

limited notion of considering only the next hop to a network the way distance-vector

algorithms like RIP function. Instead, the BGP router stores more complete information

about the path (sequence of autonomous systems) from itself to a network. Special path

attributes describe the characteristics of paths, and are used in the process of route

selection. Because of its storage of not just next-hop but path information, BGP is

sometimes called a path vector protocol.

BGP chooses routes using a deterministic algorithm that assess path attributes and

chooses an efficient route while avoiding router loops and other problem conditions. The

selection of routes by a BGP router can also be controlled through a set of BGP policies

that specify, for example, whether an AS is willing to carry traffic from other ASes or not.

However, BGP cannot guarantee the most efficient route to any destination, because it

cannot know what happens within each AS and therefore what the cost is to traverse each

AS.

BGP's operation is based on the exchange of messages that perform different functions.

BGP routers use Open messages to contact neighboring routers and establish BGP

sessions. They exchange Update messages to communicate information about reachable

networks, sending only partial information as needed. They also use Keepalive and Notifi-

cation messages to maintain sessions and inform peers of error conditions. The use of

these messages is summarized in the topic on BGP general operation, with details

(including message formats) in the section on detailed operation.

BGP uses TCP as a reliable transport protocol, so it can take advantage of the many

connection setup and maintenance features of that protocol. This also means that BGP

doesn't need to worry about issues such as message sequencing, acknowledgements, or

lost transmissions. Since unauthorized BGP messages could wreak havoc with the

operation of the Internet, BGP includes an authentication scheme for security.