Transcript Chapter 15

Routing Within An
Autonomous System
(RIP, OSPF)
Chapter 15
Static vs Dynamic Interior Routes
Interior Routers


Two routers within an autonomous system
Are considered to be interior to one another
How do they learn about their own networks?

Small, slowly changing systems
Establish and modify routes by hand
Update table when new network added or deleted
Figure…routing is trivial
Figure 15.1
Disadvantages of manual system


Cannot accommodate rapid growth
Cannot accommodate rapid change
Need automated methods



Respond to change more easily
Improve reliability
Better response to failures
Figure 15.2
Multiple physical paths




Usually pick one to be primary
Router(s) fail along primary, must change
Manually: time-consuming & error prone
Even in small internets need automated system
For automation



Interior routers must communicate
Exchange routing information
Once data established, advertise
One interior router advertises to other autonomous
systems via Exterior Gateway Protocol
No single interior protocol has emerged


Varied topologies and technologies
Tradeoffs between simplicity & functionality
Easy to install & configure; less functionality

Multiple protocols have become popular
Small AS

Choose a single one; use exclusively internally
Larger AS

Often choose a small set
Interior Gateway Protocol

Used as generic description

Refers to any algorithm used by interior routers

Routers may run BGP to advertise reachability
Need an IGP to obtain information within the AS
Figure 15.3
Routing Information Protocol (RIP)
One of most widely used IGPs


Also know as route-d
Came from Univ of CA – Berkeley
Developed for machines on their LANs


Relies on physical network broadcast to make
routing exchanges quickly
Not designed to use on large WANs
Versions of RIP adapted for WANs are sold

Popularity not only due to technical merits
Was distributed with popular 4BSD UNIX systems
Lots of TCP/IP sites used without even considering
technical aspects
Once installed it became the basis for local routing

RIP was built and adopted without a standard
Most implementations derived from Berkeley code
Interoperability was limited

Many undocumented details and subtleties
New versions led to more problems
Standard appeared in June 1988
RIP Operation


Straightforward distance-vector routing
Partitions participants into two categories
Active:


Advertise their routes
Only routers
Passive:


Do not advertise
Host must use passive mode

Active RIP routers broadcast every 30 seconds
Sends routing update message


Takes information from current routing database
Update is a set of pairs
(IP address, integer distance to that network)
Hop count is used as the distance metric
One hop: directly connected
Two hops: reachable through one other router
Hops = number of networks datagram will encounter
Hop count for shortest path not always optimal


3 Ethernets faster than 2 satellites
Some RIP implementations allow assignment of artificially
high hop counts when advertising slow networks

Active & passive routers listen to broadcasts
All update tables according to DV algorithm
May take some time for advertisements to propagate

Routers use hysteresis to improve performance
Does not replace a route with an equal cost route
Prevents oscillation among equal paths

Timers are used on all routes
Solves problem of routes through crashed routers
Start timer when install route in table
Restart whenever receive msg advertising that route
Route is invalid if 180 seconds pass without another
advertisement

RIP must handle three types of errors
Routing loops


Algorithm does not explicitly detect routing loops
Must either assume participants can be trusted or take
precautions to prevent
Instabilities



Must limit hop count to prevent
Maximum possible distance value is 16
If legitimate hop count is higher, must divide the internet into
sections or use an alternate protocol
Slow convergence



Routing messages propagate slowly across the network
Can lead to inconsistencies
Not unique to RIP; fundamental problem in any DV
protocol
Hop count limit helps but does not eliminate
Figure 15.4
Solving slow convergence

“Good news travel quickly; bad news travel
slowly”
Quick to install good route
Unreachable only after timeout; then learn and
propagate new route

Split horizon update
Router does not propagate information over the
interface from where the route arrived
R2 would not advertise route to network 1 to R1
If R1 loses connectivity, it must quit advertising
After a few rounds of routing updates, all routers
agree that the network is unreachable
Does not prevent all routing loops

Hold down
Router ignores information for a period of time
Typical time is 60 seconds
Done after receiving msg that network is unreachable
Wait so all machines can get bad news; keeps from
mistakenly accepting an out-of-date message
All machines must have same idea of hold down

Otherwise, get routing loops
Disadvantages:


If routing loop occurs, will be preserved for the hold down
Also preserves incorrect routes during the hold down time
Even when alternatives exist

Poison reverse
When connection disappears


Advertising router keeps entry for few update periods
Puts infinite cost in the broadcasts
Combine with triggered updates



Router sends immediate broadcast when get bad news
Not wait for next broadcast time
Minimizes time it is vulnerable to believing good news

These techniques solve some problems;
introduce others
Triggered updates



Suppose many routers share common network
Single broadcast changes all tables; triggering more
Broadcast avalanche
Broadcasts

Take substantial bandwidth themselves
Loops prevent stopping loops

Looping messages may prevent routing msgs to break loops
Hold down in WANs


Period so long, higher level protocol timers may expire
Breaks the connections
RIP1 message format

Messages are of two types
Routing information messages

Periodic broadcast of unsolicited response messages
Messages to request information



Routers or hosts can ask for info with request command
Routers reply using a response command
Both use same format
Figure 15.5
Command
Meaning
1
Request for partial or full routing information
2
Response (network-distance pairs from sender’s routing table)
9
Update Request (used with demand circuits)
10
Update Response (used with demand circuits)
11
Update Acknowledge (used with demand circuits)
RIP2 Address Conventions

Skip…..
RIP route interpretation and aggregation

Version 1 contains no provision for subnet mask
Originally designed for classful addressing
Extended to allow subnetting
Important restriction:



Subnet routes can only go in updates sent across networks
that are part of the subnetted prefix
Cannot use with variable-length subnet addresses or
classless
Due to not having explicit subnet mask information
May have updates for networks in & out of prefix

Router must prepare different update messages
RIP2 extensions


Contains provisions for explicit subnet mask
Also include explicit next-hop information
Prevents routing loops
Prevents slow conversion
RIP2 message format

Puts new info in unused octets of address field
Router can use both versions simultaneously
Version number in same octet; inspect before process

Adds 16-bit ROUTE TAG
Identify the origin of the route
Figure 15.6
Transmitting RIP messages

Messages do not have explicit length field
Nor any explicit count of entries

Rely on delivery mechanism to tell length
With TCP/IP


Rely on UDP to tell receiver the message length
RIP operates on UDP port 520
Disadvantage of RIP hop counts


RIP restricts routing to a hop-count metric
RIP restricts size of any internet using it
Has small hop count value for infinity (16)
Limits span to at most 15 routers between hosts
Is not a limit on total number or density of routers

In any case, hop count is a crude measure
Not always get least delay or highest capacity routes
Makes routes static; cannot change due to load
The Hello Protocol
IGP that uses routing metric other than hops




Now obsolete
Historically, used in original NSFNET backbone
“fuzzball” routers
Uses metric of delay
Provides two functions
Synchronizes clocks among a set of machines
Allows each machine to compute shortest delay paths

Messages carry timestamp as well as routing info
Each participating machine maintains table
Contains best estimate of neighboring machine clocks
Transmit timestamp with each packet

Receiver computes estimate of delay on the link by using the
timestamp and its estimate of the sender’s clock
Periodically poll neighbors to update clock estimates

Standard D-V approach for update
Send table of destinations & estimated delays
Receiver’s update tables if cheaper route advertised
Delay Metrics & Oscillation
Is delay a good routing metric?



Would seem so
Worked well in the early Internet backbone
Instability is the reason most protocols do not
use delay
Any protocol that changes routes quickly can become
unstable
Hop counts fixed; delay is not

Minor variations in delay measurements occur
Hardware clock drift
CPU load during measurement
Delays by link-level synchronization

If react quickly to slight variations, get twostage oscillation
Switch back and forth between alternate paths
Heuristics to help avoid oscillation

Hold down
Slows down changing

Round off measurements or use threshold
Ignore differences less than the threshold

Use average measurement
Keep average of recent measurements
Use K-out-of-N rule

K of the most recent N measurements must be less than the
current delay before route can be changed
Can still have instability

Due to comparing delays on paths with
different characteristics
Traffic has dramatic effect on delay
As load increases, delay grows rapidly

Fall into positive feedback cycle
Burst of traffic at one place increases delay
Protocol changes route
New traffic may cause another change in delay

Another route change occurs
Must have mechanism to dampen oscillation
Previous heuristics may not help


They help in simple case for paths with same
throughput characteristics
Not good when paths have different delay and
throughput characteristics
Compare serial line and satellite link






First, both paths idle; serial line have much less delay
Then, traffic quickly overloads low capacity line
Satellite delay will be less; change to it
High capacity; load not significantly change delay
But, unloaded serial line will now become attractive
Routing will change again and the cycle will continue
Oscillations do occur in practice
Difficult to manage
Combining RIP, Hello, and BGP
Single router may use multiple protocols

Interior Gateway Protocol
Gather routing information within AS

Exterior Gateway Protocol
Advertise routes to other ASs

Should be easy to combine the two
Technical and political obstacles exist
IGP protocols are routing protocols



RIP and HELLO used to update routing tables
Get info from other routers inside AS
routed implements RIP
Advertises information from local routing table
Updates local table when it receives updates
RIP trusts routers within the AS to send correct data
Exterior protocols (BGP) do not trust routers

Do not advertise all possible routes in local table
Keep database of reachability
Apply policy constraints when sending/receiving info


Ignoring policy constraints can make some
parts of the internet unreachable
Example:
Suppose router running RIP
Propagates route to Purdue; actually has no route
Other RIP routers will accept and update
Will pass Purdue traffic to the erroneous router
Problem if EGP protocol not have policy constraints


Border router pass illegal route to other ASs
Purdue may become unreachable for parts of the internet
Gated: Inter-AS Communication
gated


Interface between autonomous systems
Understands multiple protocols
Both IGP’s and BGP

Ensures policy constraints are honored
Can accept RIP msgs and modify local table (routed)
Can advertise routes from within AS using BGP



Has rules on which networks it may & may not advertise
Also has rules on how to report distances to those networks
Links IGP with BGP
Open SPF Protocol (OSPF)
Chapter 13: link state algorithm


Uses SPF to compute shortest paths
Scales better than distance-vector algorithms
OSPF is an IGP using link state algorithm



Designed by Internet Engineering Task Force
To encourage adoption of link state technology
Tackles several ambitious goals
Open standard

Anyone can implement without license fees
Includes type of service routing



Have multiple routes for a given destination
Choose by TOS field in IP header
OSPF first among TCP/IP protocols to have this
Provides load balancing

Distributes traffic over multiple, same cost routes
Can partition routers and networks into areas

Permits growth; makes management easier
Allows exchanges to be authenticated

Variety of authentication schemes
Supports host-specific, subnet-specific, and
classless routes
Accommodates multi-access nets (Ethernet)
Can describe network via virtual network


Abstracts away from details of physical
connections
Provides flexibility for managers
Allows routers to exchange routing info
learned from external sites

Distinguishes where information came from
Figure 15.7
Type
Meaning
1
Hello (used to test reachability)
2
Database description (topology)
3
Link status request
4
Link status update
5
Link status acknowledgement
Figure 15.8
Routers exchange database description msgs


Used to initialize network topology database
During exchange:
One router is master; other is slave
Slave acknowledges each description with a response

Topology database may be large
Can divide into several messages using I and M bits
I = 1: is initial message
M = 1: additional messages follow
Bit S indicates if sent by master (1) or slave (0)
Sequence numbers used to make sure all received
Figure 15.9
Link Type
Meaning
1
Router link
2
Network link
3
Summary link (IP network)
4
Summary link (link to border router)
5
External link (link to another site)
Link status request message

After exchanging DB descriptions, router
may discover parts of its DB are out of date
Requests neighbor to send update
Lists specific links it wants info about
Neighbor responds with most current information
about those links
Figure 15.10
Figure 15.11
Links from router to:
- given area
- specific network
- single, subnetted IP network
- networks at other sites
Figure 15.12
Routing with Partial Information
Hosts can have partial information

Rely on routers
Not all routers have complete information


Usually single router in AS connects to others
Suppose site connects to global Internet
At least one router must have connection to an ISP
Routers inside AS know all destinations within

Have default route to send all traffic to the ISP
Examine routing tables

Routers at center of Internet know all
Have complete set of routes to all destinations
Learn from the routing arbiter system
They do not use default routing
If destination address not in their table:



Either address is not valid, or
Address is valid but currently unreachable
Routers beyond those at center do not know all
Have an incomplete set of routes
Rely on default routes for addresses they do not have
Using default routes for most routers has
consequences

Local routing errors can go undetected
Machine in AS routes packet to external AS
Suppose should have gone to local router
External system will route it back
Preserves connectivity even if routing incorrect
Very bad for a WAN

Routing update messages will be smaller
A good thing!
Summary
Must have automated routing procedures
IGPs


Used by 2 routers under control of single mgr
Uses either DV or link state algorithm (SPF)
RIP: DV implemented by routed

Uses split horizon, hold-down and poison reverse
Hello: obsolete; uses delay versus hop count
OSPF: uses link status algorithm
gated

Interface between IGP (RIP) and EGP (BGP)