추천싸이트: 즐겨찾기 해놓으면 도움이 되는 사이트
추천 까페 : 네이버 네트워크 무료 강좌 추천 까페
카페운영자에게 메신져로 물어보기
네이트온 : assaman@lycos.co.kr
MSN : superljk@hotmail.com
무료상담 : 080-384-3233
Fault-Tolerant Topologies 21
00:37:06: IP: s=172.16.10.20 (Ethernet0), d=224.0.0.2, len
48, rcvd 2
00:37:06: IP: s=172.16.10.1 (local), d=224.0.0.2
(Ethernet0), len 48, sending broad/multicast
00:37:09: IP: s=172.16.10.20 (Ethernet0), d=224.0.0.2, len
48, rcvd 2
00:37:09: IP: s=172.16.10.1 (local), d=224.0.0.2
(Ethernet0), len 48, sending broad/multicast
Proxy ARP
You can configure some IP stacks to take advantage of proxy ARP. You may
recall that, under normal circumstances, workstations will use the ARP protocol to find the hardware addresses that are on their local network. When
using proxy ARP, however, these workstations will send out ARP requests
for every IP device that they want to communicate with, regardless of
whether or not it is on their local network. Any router that is hearing this
request and that is able to reach the desired IP address can respond to the
ARP with its own MAC address. From the workstation’s view, it looks like
the whole world is one big LAN. The routers take care of the details of reaching
remote segments. Proxy ARP is now enabled by default in all Cisco routers.
The end result is that workstations can dynamically locate redundant
paths out of the LAN. By sending out the proxy ARP request (which is a
broadcast), a response can come from any router able to reach the required
destination, and thus if one router fails, the workstation can immediately
begin to communicate with the internetwork through any other available
routers. Understand, however, that overhead will result on any router per
forming proxy ARP.
To configure workstations to run proxy ARP, simply set the default gateway of the workstations to their own IP address. Once you have reconfigured
your default gateway to the IP address of the workstation, try pinging a remote
device. Turn on debug ip packeton the router, and see what happens.
AppleTalk
Have you ever wondered why you don’t have to play these little gateway
games with AppleTalk? The reason is that both addressing and default
router configuration are dynamic with this protocol. With AppleTalk, the
workstations actually listen to the RTMP (Routing Table Maintenance Protocol) routing updates (this fact will become important in later chapters as
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
22 Chapter 1
Hierarchical Network Design
we specify routing protocols). They don’t build routing tables as routers do,
but they do pay attention to the source AppleTalk address of the update.
They then use that address as their default gateway! You may recall that
RTMP updates are broadcast every 10 seconds, which means that if you lose
your default router on a network, workstations will take a maximum of 10
seconds to learn any redundant router address.
Internetwork Packet eXchange (IPX)
Internetwork Packet eXchange (IPX) is also dynamic in assignment of
address and default router. Instead of listening in on IPX routing updates as
AppleTalk clients do, however, IPX clients can issue a “find network number” request. Any router that can provide access to the requested network
answers this request. If that particular router goes away, the client will automatically reissue the request. If there is a different path out, the new router
will answer the client request, and the client can then take advantage of the
new path. Once again, completely dynamic.
What this means is that at the access layer any time that you provide two
paths out, AppleTalk and IPX clients will automatically find them and use
them, and that increases the fault tolerance of your network. As we mentioned, if the clients cannot find paths out, the internetwork is not much use
to them. IP clients are typically more challenging, because they generally are
not as dynamic at finding paths out as IPX or AppleTalk clients.
Redundant WAN Connections
As you have just seen, you can provide redundancy in the links between clients and servers on the LAN using several techniques. Now we will look at
ways to provide redundancy inside the WAN.
Consider the network illustrated in Figure 1.9. This is a full mesh net
work, in which every node has a direct link to every other node. For fault tol
erance, this is great! It is far from efficient, however, and does not scale well.
Also, it has departed from our hierarchical topology that we looked at ear
lier. There is a solution, however, that will preserve hierarchy while provid
ing redundancy in the WAN.
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
Fault-Tolerant Topologies 23
FIGURE 1.9 A full mesh network
Partial Mesh Topology
We have implemented a partial mesh in the network shown in Figure 1.10.
Notice that we have preserved our hierarchy, yet each node has a redundant
link to the layer above it. This design provides all the advantages of hierarchical design, is scalable, and can take advantage of load balancing.
FIGURE 1.10 Redundant hierarchical network
You can add the additional WAN connections in several ways. You could
add them in identical pairs.that is, you could install two T1 lines rather
than one. This provides the ultimate in redundancy. If one T1 fails, another
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
24 Chapter 1
Hierarchical Network Design
is waiting to go. From a cost perspective, however, this can be similar to buying two new cars just in case one gets a flat tire. True, you will probably
never have to walk to work, but that security will certainly cost you.
An alternative to identical connections to the next layer is using links that
are not the same, that is, perhaps a T1 and a 56Kbps backup line. Should the
primary line fail, internetwork connectivity can be preserved, although generally at a reduced level. Once again, cost will most likely determine the
capacity of the backup line.
Cisco has a solution that is a special case of this second example, that is,
the two connections are not the same. In this case, the second, or backup,
line is not even running until the primary line fails! We will look at this
solution next.
Dial-on-Demand Routing (DDR) Backup
Not all redundant links have to be dedicated lines. In many cases, an ISDN
BRI (Basic Rate Interface) is used to back up a dedicated leased line. This can
be a great advantage, because you will probably not want to bring the ISDN
up unless the primary line fails (or becomes overloaded). Cisco’s DDR
allows this configuration. The ISDN line can be configured to become active
only when the primary line either fails or is under heavy load. Of course,
should the primary line fail and you have to depend on your backup, you will
likely not have your normal bandwidth available. You will, however, likely
be paying significantly less than you would to have a pair of dedicated lines.
To configure a BRI interface as a backup link or to relieve overloaded
lines, use the following commands:
Router#config t
Enter configuration commands, one per line. End with
CNTL/Z.
Router(config)#int s0
Router(config-if)#backup ?
delay Delays before backup line up or down
transitions
interface Configure an interface as a backup
load Load thresholds for line up or down
transitions
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
Fault-Tolerant Topologies 25
Router(config-if)#backup int bri0
Router(config-if)#backup delay ?
<0-4294967294> Seconds
never Never activate the backup line
Router(config-if)#backup delay 60 ?
<0-4294967294> Seconds
never Never deactivate the backup line
Router(config-if)#backup delay 60 90
Router(config-if)#backup load ?
<0-100> Percentage
never Never activate the backup line
Router(config-if)#backup load 60 ?
<0-100> Percentage
never Never deactivate the backup line
Router(config-if)#backup load 60 30
The commands above set BRI0 as both a backup link and as a load-sharing link. If serial 0 goes down, the BRI will wait 60 seconds to give the line
a chance to recover. If, after 60 seconds, the line does not come up, the BRI
will dial. It will not disconnect until after serial 0 has been up consistently for
90 seconds. The backup load command tells the BRI to dial up and share
the load with the serial 0 interface if the load on the line reaches 60 percent
saturation. The BRI will disconnect when the load drops below 30 percent.
Performance: Load Balancing
Redundant links are not cheap to operate, but they are called for in some situations. If you are going to pay for redundant links, you would likely want
to use both lines when they are both available, and that brings us to load balancing.
A good design rule is to keep bandwidth consistent within a layer of hierarchy whenever possible and to use technologies such as DDR when purchasing
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
(www.sybex.com) www.sybex.com
26 Chapter 1
Hierarchical Network Design
equal links is not possible. This will avoid the pinhole congestion issue and
generally facilitates predictability within the network. Of course, this is not
always possible, especially when the redundant paths are expensive leased
lines!
Internet Protocol (IP)
FIGURE 1.11
With most IP routing protocols, load balancing is automatic. Dynamic routing protocols are supposed to find the redundant paths, and dynamic IP routing
protocols will use both available paths. This can, however, not always be a
good thing.
Pinhole Congestion
Difficulties can arise when the multiple paths out do not have the same bandwidth or cost. Suppose that you have a T1 and a 56Kbps line (for backup)
connecting your access-layer router into distribution layer routers, as shown
in Figure 1.11.
Pinhole congestion
56Kbps
T-1
Access Router
Some routing protocols (for example, those that use hop count) could see
these two paths and load balance across them just fine until the 56Kbps line
is full. At that point, the load is equally balanced. These protocols, however,
are not smart enough to realize that more than 90 percent of the total bandwidth is going unused on the T1! Once any link is operating at capacity,
these routing protocols are not capable of sending additional traffic across
links still not at capacity, because they do not understand capacity as a metric.
This problem is called pinhole congestion, which you can avoid by using
advanced routing protocols such as Enhanced IGRP.
Internetwork Packet eXchange (IPX)
By default, IPX will not load balance across multiple links; however, Cisco
provides a way to enable IPX load balancing. You can use the ipx
maximum-paths command, which specifies a number of links to load balance
across.
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
(www.sybex.com) www.sybex.com
Fault-Tolerant Topologies 27
Use the command ipx m?to access all the commands that start with ipx
m. Notice that the ipx maximum-pathsand ipx maximum-hopsare the commands listed. If you change the default parameters on one router, you need
to change these parameters on all your routers.
The command ipx maximum-pathstells IPX to consider that there might
be more than one link to the same location. By default, IPX will not consider
that a second link could exist and will not provide a round-robin load balance. The ipx maximum-paths command solves this problem.
Router(config)#ipx m?
maximum-hops maximum-paths
Router(config)#ipx maximum-paths ?
<1-64> Number of paths
IPX RIP uses only 15 hops by default; it will discard any packet when it
reaches 16 hops. If your internetwork grows beyond 15 hops, you need to
configure the ipx maximum-hops command on all your routers.
Router(config)#ipx maximum-hops ?
<16-254> Max hops
AppleTalk
AppleTalk, like IPX, considers only one path to a remote network. You can
set the number of parallel routing paths that can be used by AppleTalk by
using the appletalk maximum-paths command. Remember to set this on
all your routers, not just one router.
Router(config)#apple m?
macip maximum-paths
Router(config)#appletalk maximum-paths ?
<1-16> Number of parallel routing paths
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
28 Chapter 1 Hierarchical Network Design
Summary
Network topology design can make the rest of the design process sig
nificantly easier or more difficult. Cisco recommends using hierarchical
design, which offers many benefits including:
Predictability
Scalability
Efficiency
Cost control
Security
Further, Cisco recommends that small- to medium-size businesses use a
three-layered approach to hierarchy, consisting of these layers:
Core
Distribution
Access
Each layer has clearly defined functions, and once the network is established, it can scale significantly before it needs to be reengineered.
Topologies that enhance network fault-tolerance are often required. IPX
and AppleTalk dynamically find their gateways to the internetwork, but for
IP features such as HSRP and proxy, ARP can improve fault tolerance in the
workstation to router communication. Redundant WAN links can provide
additional fault tolerance and can be used inside hierarchical designs. Tech
nologies such as DDR provide for backup links. When redundant links are
used, design consideration should be given to load balancing. Identify and
avoid issues such as pinhole congestion.
Key Terms
Before you take the exam, be certain you are familiar with the following
terms:
access layer
core layer
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
Summary 29
data compression
distribution layer
header
hierarchy
load balancing
phantom router
route summarization
snapshot routing
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
30 Chapter 1 Hierarchical Network Design
Review Questions
1. Which of the following are advantages of hierarchical design?
A. Fault tolerance
B. Scalability
C. Ease of management
D. Predictability
E. All the above
2. Which of the following are layers in Cisco’s three-layer hierarchical
design?
A. Backbone
B. Core
C. End node
D. Access
E. Distribution
3. Which of the following should be included at the core layer?
A. Packet filtering
B. Firewalling
C. Fast throughput
D. Fault tolerance
E. Additional devices
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
Review Questions 31
4. How many layers of hierarchy should you add below the access layer?
A. None
B. One
C. Two
D. Three
E. Four
5. Which of the following are permitted at the distribution layer?
(Choose all that apply.)
A. Packet filtering
B. Access lists
C. Queuing
D. Redundant WAN connections
E. Firewalls
6. Which of the following protocols allow for dynamic location of
default routers?
A. IP
B. IPX
C. AppleTalk
D. NetBEUI
7. Which command do you use to configure HSRP on a Cisco router?
A. Router#hsrp 0 172.16.20.1
B. Router(config)#ip hsrp 172.16.20.1
C. Router(config)#ip standby 172.16.20.1
D. Router(config)#standby ip 172.16.20.1
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
32 Chapter 1 Hierarchical Network Design
8. Which of the following methods will allow IP workstations to locate
routers dynamically?
A. HSRP
B. Workstation listening to routing protocols
C. Router location request
D. Proxy ARP
E. RTMP
9. You need to add a new site to your hierarchical network. Which of the
following are possible places to connect the new site into your existing
network?
A. Access layer
B. Distribution layer
C. Core layer
D. Backbone
10. When designing fault-tolerant network topologies, which of the
following can DDR accomplish?
A. Back up a primary link in case of failure.
B. Promote a router from access to distribution layer.
C. Populate Enhanced IGRP tables with routing information.
D. Back up a primary link in case of heavy network load.
11. Select the topology in which it is easiest to troubleshoot connectivity
issues.
A. Bus
B. Ring
C. Hierarchy
D. Mesh
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
Review Questions 33
12. If you need to route IPX over 19 hops, what command do you use?
A. ipx maximium-hops 19
B. maximum-paths 19
C. ipx maximum-paths 19
D. maximum-hops 19
13. You have a T1 link from an access-layer router to a distribution-layer
router, and you have a BRI DDR connection to another distribution-
layer router. The DDR is configured to run in case of failure. Which
of the following do you have?
A. Proxy ARP
B. Fault tolerance
C. Load balancing
D. HSRP
E. None of the above
14. What is the problem caused during IP load balancing by routing
protocols that use hop count as a metric?
A. Pinhole congestion
B. Failure
C. Convergence delay
D. You can’t load balance IP.
15. Which of the following best describe a use for DDR?
A. Backup link in case of primary link failure
B. Additional link used for load balancing
C. Backup link in case of excessive network traffic
D. Additional default gateway for IP clients
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
34 Chapter 1 Hierarchical Network Design
16. Which of the following best describes the function of proxy ARP?
A. The host pings the destination site to discover which router to use.
B. The host uses the ARP protocol to get the router’s IP address so
that it can find the router’s MAC address.
C. The host uses the ARP protocol to get the router’s destination’s
hardware address, and the router responds with its (the router’s)
MAC address.
D. The router uses the ARP protocol to get the host’s hardware
address to see if it needs to communicate with the internetwork.
17. You have a T1 link from an access-layer router to a distribution-layer
router, and you have a BRI DDR connection to another access-
layer router. The DDR is configured to run in case of failure. Which
of the following do you have?
A. Proxy ARP
B. Fault tolerance
C. Load balancing
D. HSRP
E. None of the above
18. How can redundant links be added into a hierarchical design without
breaking the hierarchy?
A. Full mesh
B. Partial mesh
C. Create a ring
D. They can’t
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA
Review Questions 35
19. Your customer has a hierarchical network design. Redundancy and
reliability are the most important at which layer?
A. Backbone
B. Distribution
C. Access
D. Core
20. Which of the following usually should be supported at the distribution
layer?
A. DDR and Ethernet switching
B. Creation of separate collision domains (segmentation)
C. Routing between VLANs and other workgroup support functions
D. User and workgroup access to internetwork resources
(www.sybex.com) www.sybex.com
(Copyright ⓒ2001 SYBEX , Inc., Alameda, CA) Copyright ⓒ2001 SYBEX , Inc., Alameda, CA