67
1 Jim Binkley An Overview of Routing Theory IP Routing Jim Binkley Portland State University

An Overview of Routing Theory Jim Binkleyweb.cecs.pdx.edu/~jrb/routing/lectures/pdfs/ROUTTHE2.pdf · An Overview of Routing Theory Jim Binkley ... this could be the model for an intranet

  • Upload
    lamkiet

  • View
    217

  • Download
    0

Embed Size (px)

Citation preview

1Ji

m B

inkl

ey

An

Ove

rvie

w o

f Rou

ting

Theo

ry

IP R

outin

gJi

m B

inkl

eyPo

rtlan

d St

ate

Uni

vers

ity

2Ji

m B

inkl

ey

Rou

ting

Theo

ry

topo

logi

es a

nd sc

alab

ility

basi

c to

ols &

idea

s & a

ttrib

utes

»st

atic

vs d

ynam

ic, f

lood

ing,

tunn

els,

cont

rol t

heor

y

som

e is

sues

; e.g

., co

nges

tion

algo

rithm

s: v

ecto

r-di

stan

ce v

s lin

k-st

ate

3Ji

m B

inkl

ey

fund

amen

tal i

deas

rout

ing

-fin

ding

a p

ath

from

one

end

to th

e ot

her f

or a

pac

ket

we

need

one

or m

ore

algo

rith

mst

hat a

re

mos

t lik

ely

dist

ribu

ted

amon

gst a

set o

f ho

sts a

nd ro

uter

wha

t are

the

prop

ertie

s of s

aid

algo

rithm

?w

hat i

ssue

s aff

ect i

t?

4Ji

m B

inkl

ey

elem

ents

of a

rout

ing

sche

me

rout

ing

prot

ocol

s tha

t allo

w in

fo to

be

gath

ered

and

dis

tribu

ted

-ro

utin

g ag

ents

co

mm

unic

ate

with

thes

e pr

otoc

ols

rout

ing

algo

rith

ms -

may

be

dist

ribut

ed,

use

prot

ocol

s and

dat

a to

det

erm

ine

and

diss

emin

ate

path

sro

utin

g da

taba

ses (

tabl

es in

rout

ers)

(to b

oard

wal

k, v

ia n

ew je

rsey

, $1

00)

5Ji

m B

inkl

ey

a ro

utin

g do

mai

na

rout

ing

dom

ain

== se

t of r

oute

rs u

nder

sam

e ad

min

runn

ing

sam

e ro

utin

g pr

otoc

ol

e.g.

, all

thes

e ro

uter

s are

con

trolle

d by

Joe

Bob

Inc,

run

OSP

F

6Ji

m B

inkl

ey

idea

l rou

ting

algo

rithm

corr

ect -

wha

t if a

lgor

ithm

s too

com

plex

?ro

bust

-ca

n de

al w

ith ro

uter

rebo

ot?

–tw

o pr

oble

ms:

loss

of r

oute

r and

loss

of l

ink

(i/f f

ailu

re ..

.)

stab

le -

do ro

utin

g ch

ange

s sta

biliz

e in

dis

tribu

ted

syst

em?

–co

nver

genc

e-a

stat

e, a

ll ro

uter

s hav

e “s

ame”

rout

ing

tabl

e

effic

ient

-al

l rou

ting

and

no d

ata

not g

ood

topo

logi

cally

flex

ible

, can

allo

w a

ggre

gatio

nm

aint

aina

ble

-adm

in n

ot to

o co

mpl

icat

edsc

alab

leto

man

y ro

uter

s, m

any

host

s? (d

istri

bute

d)de

adlo

ck?

-ro

utin

g lo

op-f

ree?

secu

re -

can

intru

ders

inje

ct ro

utes

?

7Ji

m B

inkl

ey

clas

sic

prob

lem

rout

ing

loop

path

to Z

?“h

ey, I

’m Z

...”

crea

ted

stat

ical

ly o

r dyn

amic

ally

AB

C

(To

Z, V

ia C

)

(To

Z, V

ia A

)

(To

Z, V

ia B

)

8Ji

m B

inkl

ey

topo

logy

Tane

nbau

mm

entio

ns lo

gica

l abs

urdi

ties

no r

oute

r -e

very

hos

t wire

d to

eve

ry o

ther

ho

st (

mes

h )

N *

N w

ires,

go

ahea

d ad

d a

host

...1

rout

er -

for a

ll ho

sts (

star

)–

1 he

ck o

f a ro

utin

g ta

ble

–th

is is

why

cor

e ca

n ha

ve p

robl

ems

–si

ngle

poi

nt o

f fai

lure

?!–

ever

ythi

ng sh

ould

be

fully

con

nect

ed?!

9Ji

m B

inkl

ey

one

solu

tion

typi

cally

a fl

atte

ned

tree

to g

ive

hier

arch

yat

the

top

a sm

all c

ircle

of c

ore

rout

ers t

hat

know

all

the

rout

esid

ea:

defa

ult r

oute

–if

you

are

not i

n th

e ce

nter

AN

D y

ou d

on’t

know

wha

t to

do w

ith it

, se

nd it

“U

P”to

sm

arte

r ent

ity

10Ji

m B

inkl

ey

rout

ing

hier

arch

y

R

RR

R

net

Leve

l 1

Leve

l 2

defa

ult r

oute

H

core

rout

ers

R

rout

er

this

cou

ld b

e th

e m

odel

for a

n in

trane

t OR

the

Inte

rnet

as a

who

le (m

ultip

le c

ore

rings

)

11Ji

m B

inkl

ey

leve

ls

the

top

leve

l has

to k

now

eve

ryth

ing

–in

a ro

utin

g do

mai

n, ro

uter

s kno

w a

ll th

e ro

uter

s at l

evel

1–

in B

GP,

top

of In

et, r

oute

rs m

ay h

ave

full

rout

ing

tabl

e ro

uter

s may

als

o ch

oose

to h

ide

inte

rior

stru

ctur

e (s

ubne

t/CID

R) i

n or

der t

o ag

greg

ate

rout

ing

info

–re

duce

rout

ing

tabl

e si

ze e

lsew

here

12Ji

m B

inkl

ey

divi

de ro

utin

g w

orld

into

3 p

arts

topo

logy

IETF

ISO

/OSI

sam

e “l

ink”

or

wire

none

, int

ra-li

nk?

none

, int

ra-li

nk?

ente

rpris

e or

cam

pus

Inte

rior G

atew

ayPr

otoc

ol -

IGP

intra

-dom

ain

rout

ing

prot

ocol

betw

een

ente

rpris

esEx

terio

r Gat

eway

Prot

ocol

- EG

Pin

ter-

dom

ain

13Ji

m B

inkl

ey

prot

ocol

s acc

. to

topo

logy

topo

logy

IETF

ISO

/OSI

intra

-link

AR

PES

-IS

intra

-dom

ain

RIP

, RIP

(2),

OSP

FIS

-IS

inte

r-do

mai

nEG

P, B

GP(

4)ID

PR

14Ji

m B

inkl

ey

the

Inte

rior -

RIP

or O

SPF

out

Joe

Bob

Inc’

sN

etw

ork

Map

link

neig

hbor

sor

pee

rs

the

“wor

ld”

(fro

m ro

uter

PO

V)

15Ji

m B

inkl

ey

scal

abili

ty is

a L

AR

GE

conc

ern

10’s

of h

osts

and

a fe

w ro

uter

s -st

atic

ro

utin

gw

hat a

bout

put

ting

all o

f tho

se to

aste

rs o

n th

e In

tern

et?

(wha

t if t

he n

et/h

osts

big

ger?

)on

e ho

me/

offic

e, to

bus

ines

s/en

terp

rise,

to

stat

e, n

atio

n, p

lane

t, so

lar s

yste

m, g

alax

y...

com

pone

nts a

ffec

ted

incl

ude

the

netw

ork

addr

ess a

nd th

e ro

uter

hie

rarc

hy

16Ji

m B

inkl

ey

scal

abili

ty...

ip’s

curr

ent p

robl

ems

–ne

t/hos

ts v

ia c

lass

or e

ven

subn

et d

on’t

mat

ch n

umbe

r of

hos

ts re

ally

util

ized

–to

o m

any

rout

es in

cor

e ta

bles

(aga

in tr

ue p

ost C

IDR

)–

ip a

ddre

ss a

lloca

tion

from

cla

ss C

slic

e of

pie

mea

ns in

ef

fect

maj

ority

of n

umbe

rs a

re w

aste

d

scal

abili

ty a

ffec

ts a

ddre

sses

and

how

they

are

sl

iced

up;

als

o ro

uter

hie

rarc

hyho

w m

uch

low

-leve

l inf

o (e

.g.,

link-

leve

l det

ails

) ca

n w

e af

ford

to d

isse

min

ate

upw

ards

?

17Ji

m B

inkl

ey

Som

e cu

rren

t BG

P is

sues

# en

tries

> 1

0000

0, g

row

ing

fast

er th

an a

lin

ear r

ate?

CID

R h

elp

in m

id-9

0s h

as w

orn

out?

# en

tries

that

are

/24

or b

igge

r?–

I.e.,

not e

noug

h ag

greg

atio

n –

entri

es m

ay n

eed

dam

peni

ngag

grav

ated

by

mes

h re

dund

ancy

–if

mul

ti-ho

med

mor

e pa

ths a

nnou

nced

to w

orld

18Ji

m B

inkl

ey

N**

2 is

not

our

frie

ndH

as a

tend

ency

to sh

ow u

p in

rout

ing

prot

ocol

s/th

eory

bec

ause

nat

ural

ly:

rout

ers (

node

s) ti

mes

link

s (co

nnec

tions

)re

mem

ber A

ndre

w T

. and

wha

t hap

pens

if a

ll no

des d

irect

ly c

onne

ct!

The

prob

lem

her

e is

mem

ory

stor

age

in in

divi

dual

co

re ro

uter

s–

how

muc

h m

emor

y if

a co

re ro

uter

mus

t rem

embe

r: 1.

A

ll lin

k m

etric

s, 2.

All

host

attr

ibut

es,

3. H

ost X

hos

t at

tribu

tes

19Ji

m B

inkl

ey

to in

crea

se sc

alab

ility

, add

a

pref

ix dom

ain

rout

ers a

s 1st

ass

umpt

ion

-ass

ume

they

kn

ow a

ll th

e ro

utes

. if

that

isn’

t sca

labl

e, th

en a

dd n

ew h

iera

rchy

and

in

trodu

ce n

ew la

yer o

f stru

ctur

e, fi

nd p

oint

s of

aggr

egat

ion

IP is

mov

ing

tow

ards

this

(CID

R)

phon

e co

mpa

nies

hav

e pr

efix

es (n

ot e

noug

h...

nobo

dy is

per

fect

)N

AT,

MIP

(tunn

els)

, IPv

4+A

S, a

ll w

ays o

f doi

ng

this

(to

say

noth

ing

of 1

28bi

t add

ress

es)

20Ji

m B

inkl

ey

hier

arch

y pi

ctur

e

boun

dary

rout

er1 2

3

core

(4)

[ net

regi

on =

1..4

,su

bnet

in re

gion

,ho

st ]

Add

ress

To 3

.x.y

, via

4.1

.1

21Ji

m B

inkl

ey

gene

ral s

olut

ion

to p

robl

em e

xist

s

whe

n in

dou

bt, a

dd a

new

pre

fix to

ad

dres

s and

a n

ew S

MA

LLER

cen

ter t

o th

e w

orld

pref

ix m

usts

umm

ariz

e in

tern

al st

ruct

ure

divi

de w

orld

into

ce

nter

: (la

yer 1

) bou

ndar

y ro

uter

sdo

mai

ns: (

laye

r 2) i

nsid

e ro

uter

sbo

unda

ry ro

uter

s hav

e su

mm

ary

rout

es in

them

, no

t all.

22Ji

m B

inkl

ey

btw

, thi

s is r

elat

ed to

a m

agic

al

arch

itect

ural

des

ign

prin

cipl

ead

d a

laye

r of

indi

rect

ion

== P

OW

ERco

nsid

er IP

v6 a

ddre

ss (i

p pa

rt, IE

EE m

acpa

rt)M

obile

-IP

2-pa

rt A

WA

Y a

ddre

ss (F

A C

are-

Of-

Add

ress

, M

obile

-IP

Hom

e A

ddre

ss)

–iro

ny h

ere

is th

e po

wer

of t

his i

s not

wel

l und

erst

ood

BG

P IP

v4 “

addr

ess”

is (

set o

f AS

#s, I

P ds

t)ah

em: N

AT

sinc

e th

is m

ay a

dd to

the

IP a

ddre

ss

spac

e po

ol (

a fe

w re

al ip

add

r, m

appe

d to

man

y pr

ivat

e ip

add

ress

), un

der s

ever

e co

nstra

ints

23Ji

m B

inkl

eytrans

it A

Stra

nsit

trans

it

a st

ub A

S

to a

noth

er st

ub exch

ange

anot

her s

tub

Inet

has

mul

tiple

cen

ters tra

nsit

-pac

kets

mos

tlycr

oss

stub

-pa

cket

s mos

tly si

nkor

sour

ce (E

S)

24Ji

m B

inkl

ey

3 w

ays r

oute

s get

into

rout

ing

tabl

es stat

ic -

load

ed b

y ne

twor

k ad

min

s–

put i

nto

conf

igto

aut

o-lo

ad a

t boo

t–

rout

e ad

d 10

.0.0

.0 -g

wga

tew

ay-ip

dyna

mic

rout

ing

–do

ne b

y ro

utin

g da

emon

s run

ning

a ro

utin

g pr

otoc

olkn

eeje

rkm

odifi

catio

n to

hos

t rou

tes b

y IC

MP

redi

rect

s–

typi

cally

way

to p

opul

ate

rout

ing

tabl

e w

ith

defa

ult r

oute

25Ji

m B

inkl

ey

stat

ic ro

utin

gst

atic

pro

s:–

sim

ple,

may

be

easi

est t

hing

to d

o in

sim

ple

topo

logy

, es

peci

ally

for l

eaf h

ost w

ith 1

rout

er o

nly

–yo

u m

ay b

e sm

arte

r tha

n th

e ro

uter

s (w

ant a

pat

h th

ey

won

’t gi

ve y

ou) -

polic

y ro

utin

g po

ssib

le

stat

ic c

ons:

–ca

n’t r

eact

dyn

amic

ally

to c

rash

ed ro

uter

(stil

l nee

d tw

o ro

uter

s to

reac

t tho

ugh.

..), a

nti-r

edun

danc

y–

not s

cala

ble,

not

goo

d if

you

have

100

rout

ers t

o co

nfig

ure

right

now

(or 1

000

host

s)

26Ji

m B

inkl

ey

dyna

mic

rout

ing

we

need

a d

istri

bute

d ro

utin

g pr

otoc

olgo

als:

–in

telli

gent

(sic

?) c

hoic

e of

rout

e ba

sed

on

met

ric (s

ingl

e m

ore

likel

y th

an m

ultip

le)

–au

tom

atic

pop

ulat

ion

of ro

utin

g ta

ble

to a

void

m

anua

l set

up–

redu

ndan

cy o

f rou

ting

path

; i.e

., co

nver

genc

e po

st p

ath

failu

re d

ue to

loss

of r

oute

r, in

terf

ace,

or

sudd

en o

nset

of b

ackh

oe

27Ji

m B

inkl

ey

theo

ry: r

outin

g pr

otoc

ol ty

pes

cent

raliz

edve

rsus

dis

trib

uted

vers

us e

nd-

node

–ce

nter

is n

eces

sary

, can

’t ha

ve d

efau

lts

ever

ywhe

re»

why

not

just

com

pute

and

put

in c

ente

r?

–if

end

node

s hav

e ro

ute

info

, and

big

net

, »

not s

cala

ble,

then

rout

e ta

bles

hug

e -n

eed

to li

mit

size

som

e so

rt of

dis

tribu

ted

syst

em is

co

nven

tiona

l ans

wer

28Ji

m B

inkl

ey

type

s/to

ols -

sour

ce ro

utin

g

sour

ce-r

outin

g vs

hop-

by-h

op–

end

node

has

exa

ct P

ATH

and

dat

agra

m fo

llow

s tha

t pa

th:

( firs

t to

Joe,

the

n to

Bob

, the

n to

Gra

ndm

a’s )

–IP

opt

ion,

but

rare

ly u

sed

(stri

ct/lo

ose)

–ch

alle

nge

to se

curit

y: w

hat i

f hos

tile

entit

y co

nvin

ces

you

to ro

ute

all p

acke

ts th

roug

h it?

»m

aybe

it c

an m

asqu

erad

e as

you

afte

r tha

t?–

still

a p

ossi

ble

tool

-use

d in

BG

P, c

an c

omm

unic

ate

POLI

CIE

S, f

rom

Nov

ell t

o In

tel,

plea

se S

kip

Bel

levu

e–

as sc

alab

le a

s hop

by

hop

?

29Ji

m B

inkl

ey

rout

ing

tunn

els a

re b

asic

tool

tunn

els a

re e

xam

ple

of so

urce

rout

ing

(doh

!)–

stric

t or l

oose

?

IP d

atag

ram

with

IP h

eade

r (IP

IP)

–M

BO

NE

DV

MR

P, M

obile

-IP

(uni

cast

)

IPX

, APP

LETA

LK d

atag

ram

insi

de IP

hea

der

–C

isco

GR

E tu

nnel

s des

igne

d as

gen

eral

way

to d

o tu

nnel

s (G

ENER

IC R

OU

TIN

G E

NC

APS

ULA

TIO

N)

–IP

v6,

IPv4

IPv6

, or

IPv6

IPv4

secu

rity

tunn

els,

IP IP

SEC

IP, b

asis

of V

PNs

30Ji

m B

inkl

ey

enca

psul

atio

n sc

hem

e

IP sr

c =

rout

er tu

nnel

src

IP d

st =

rout

er tu

nnel

sink

enca

psul

ated

IPda

tagr

am (o

r IPX

or ?

??)

IPIP

(btw

: IP

next

pro

to =

= 4

for I

PIP)

IP

T

CP

data

31Ji

m B

inkl

ey

IP tu

nnel

ing

mes

h

IPv6

net

wor

k A

IPv6

net

wor

k B

IPv4

net

/tunn

el

rout

e to

net

B, f

orw

ard

to A

bor

der r

oute

r, w

hich

will

enc

apsu

late

and

send

to B

bor

der r

oute

r

rout

ers h

ere

do n

ot sp

eak

IPv6

IPv4

| IP

v6 d

atag

ram

virt

ual n

etw

ork

32Ji

m B

inkl

ey

cont

rol-t

heor

yro

utin

g is

con

trol

, with

out i

t, da

ta c

anno

t flo

wch

icke

n&eg

g pr

oble

m: r

outin

g in

term

s of

addr

essa

bilit

ym

ust e

xist

a p

riori

befo

re ro

utin

g ca

n oc

cur

–I c

an’t

send

you

pac

kets

if I

can’

t fin

d yo

u ...

–th

ink

abou

t thi

s in

term

s of e

very

rout

ing

prot

ocol

–ad

dres

sabi

lity

is fu

ndam

enta

lly li

nk d

epen

dent

–so

met

imes

we

rely

on

man

ual c

onfig

urat

ion

(ser

ial)

–so

met

imes

info

rmat

ion

auto

mat

ical

ly g

athe

red

(arp

or

mul

ticas

t on

ethe

rnet

)

33Ji

m B

inkl

ey

addr

essa

bilit

y pr

oble

m

broa

dcas

t lin

k

rout

er A

B

rout

er C

how

can

A k

now

of B

/C?

seria

l lin

kAB

and

send

its p

eers

rout

ing

data

?

34Ji

m B

inkl

ey

type

s/to

ols -

flood

ing

flood

ing

-ass

ume

N in

terf

aces

–pa

cket

com

es in

N(1

)–

pack

et g

oes o

ut N

(2)..

N(N

)

35Ji

m B

inkl

ey

be c

aref

ul w

ith fl

oodi

ng...

36Ji

m B

inkl

ey

flood

ing,

con

t.im

porta

nt ro

utin

g al

gorit

hm “

tool

”-u

sed

in

man

y ro

utin

g al

gorit

hms i

n so

me

sens

est

rong

pro

and

con

/spr

o -p

erfe

ct r

outin

g, y

ou fo

llow

the

best

pa

th (r

edun

danc

y he

re is

a fe

atur

e!)

con

-“pe

rfec

t con

gest

ion”

-you

use

up

too

muc

h ba

ndw

idth

con

–yo

u m

ay lo

op p

acke

ts

37Ji

m B

inkl

ey

flood

ing

may

be

cons

train

ed o

f co

urse yo

u m

ight

sim

ply

use

broa

dcas

t to

send

info

you

may

pas

s jud

gem

ent o

n th

e in

fo b

efor

e yo

u se

nd it

out

the

“oth

er”

inte

rfac

es–

not f

orw

ard

it if

it is

not

new

–or

inte

rest

ing

man

y ro

utin

g pr

otoc

ols r

ely

on so

me

sort

of

flood

ing

(eve

n B

GP,

alth

ough

not

link

-la

yer)

38Ji

m B

inkl

ey

rout

ing

is 2

1-w

ay p

robl

ems

path

bet

wee

n A

and

Z m

ay n

ot b

e th

e sa

me

asym

met

ric ro

utin

g N

OT

unus

ual

cons

ider

Mob

ile-I

P–

solv

es p

robl

em o

f pac

kets

TO

rem

ote

host

»pa

cket

s tun

nele

d fr

om H

OM

E to

AW

AY

–no

t pac

kets

FR

OM

rem

ote

host

»or

dina

ry “

defa

ult”

rout

ing

cons

ider

mul

ti-ho

med

stub

with

def

ault

rout

e

39Ji

m B

inkl

ey

Mob

ile-I

P as

one

-way

IPIP

tu

nnel

You

r hom

esu

bnet

pk

ts to

you

com

e to

your

hom

e su

bnet

you

mov

ed ..

.

IPIP

tunn

el“F

orei

gn A

gent

pkts

FR

OM

node

are

del

iver

edno

rmal

ly

40Ji

m B

inkl

ey

asym

met

ric p

aths

my

rout

ing

dom

ain

the

wild

s of t

he In

et ..

..

mul

ti-ho

med

stub

with

def

ault

rout

e

defa

ult

rout

e

pkts

in

41Ji

m B

inkl

ey

you

may

hav

e to

deb

ug tw

ice

this

pat

h w

orks

...

rout

ing

failu

re o

n pa

th b

ack

...

42Ji

m B

inkl

ey

cont

rol i

dea:

rout

es v

s dat

a

on a

giv

en li

nk, y

ou m

ight

SEN

D ro

utin

g in

fo, w

hich

will

cau

seda

ta to

CO

ME

BA

CK

on

that

pat

hro

utin

g in

fo O

UT

driv

es d

ata

com

ing

IN

here

I am

, ta

lk to

me!

!!da

ta fl

ows o

ppos

ite o

f con

trol

43Ji

m B

inkl

ey

exam

ple:

mul

ti-ho

med

stub

, 2

defa

ult r

oute

s

assu

me

full

rout

ing

tabl

eB

GP

peer

BG

P pe

er

inte

rior I

GP/

rout

ing

dom

ain,

lots

of i

nter

ior r

oute

rs (o

r 1)

defa

ult r

oute

+

defa

ult

rout

etra

ffic

...

44Ji

m B

inkl

ey

issu

es -

cong

estio

ndo

es n

ot re

fer t

o br

onch

ial c

ondi

tion

conn

ectio

nles

s rou

ters

hav

e on

ly so

man

y bu

ffer

s, to

o m

any

pack

ets,

they

dro

p th

emth

ings

get

wor

se a

t the

“fr

eew

ay e

xcha

nges

”do

es ro

utin

g pr

otoc

ol a

dd c

onge

stio

n bu

rden

?ho

w d

o w

e pr

even

t/det

ect c

onge

stio

n?ob

viou

sly

circ

uit-s

witc

hes d

on’t

have

this

pr

oble

m o

nce

circ

uit i

s set

, but

they

was

te

band

wid

th

45Ji

m B

inkl

ey

cong

estio

n?w

ays t

o pr

even

t con

gest

ion:

–ad

d ca

rryi

ng c

apac

ity -

big

pipe

theo

ry–

shut

up, e

spec

ially

if h

igh-

volu

me

src

how

do

we

notif

y ne

twor

k ab

out i

t?–

TCP

dete

cts c

onge

stio

n w

hen

send

er n

otes

that

AC

KS

are

mis

sing

, slo

w, o

r dup

licat

ed,

send

er sl

ows r

ate

of

send

ing,

this

is E

ND

to E

ND

met

hod,

rout

ers n

ot

invo

lved

(Dav

id C

lark

/MIT

, end

to e

nd p

aper

)–

som

e sc

hem

es h

ave

rout

ers f

orw

ard

or p

ass b

ack

cong

estio

n bi

ts

46Ji

m B

inkl

ey

cong

estio

n sc

hem

es -

rout

ers

invo

lved

IP se

nds b

ack

ICM

P so

urce

que

nch

mes

sage

to se

nder

(dep

reca

ted)

–pr

o: y

ou se

nt it

the

right

way

–co

n: y

ou p

oure

d ga

s on

the

fire

ISO

CN

LP se

ts fl

ag in

net

wor

k he

ader

–pr

o: d

oesn

’t ad

d da

ta to

net

–co

n: c

onge

stio

n no

tific

atio

n is

sent

to th

e D

ESTI

NA

TIO

N (o

h, g

oody

dest

inat

ion

calls

send

er w

ith P

OTS

? “s

hutu

p”

47Ji

m B

inkl

ey

cong

estio

n co

nt.

TCP

solu

tion

is n

ot b

ad B

UT

–w

hat a

bout

pro

toco

ls th

at u

se In

tern

et th

at

don’

t im

plem

ent o

r can

’t se

nsib

ly im

plem

ent

it? »N

FS o

r Nov

ell c

ould

but

don

’t, d

rive

TCP

out

»au

dio/

vide

o tra

nsm

issi

on is

stea

dy-s

tate

dat

a flo

w

cong

estio

n de

tect

ion

is a

n op

en q

uest

ion

larg

e is

sue

in Q

OS

term

s

48Ji

m B

inkl

ey

issu

es -

link

cost

s

we

need

a m

etri

c, w

hich

one

?–

cost

? n

ot a

ppro

pria

te w

ithin

dom

ain

but

betw

een;

e.g

., w

hich

long

-dis

tanc

e co

mpa

ny?

–ho

p co

unt-

how

man

y ro

uter

s do

we

trave

rse

–av

aila

ble

band

wid

th -

go le

ast c

onge

sted

rout

e–

spee

d of

und

erly

ing

netw

ork,

use

ATM

as

oppo

sed

to 1

200

baud

mod

em?

–tim

e: sh

orte

st p

ath

in te

rms o

f tim

time

is ty

pica

lly a

n ap

p ...

49Ji

m B

inkl

ey

issu

es -

link

cost

s (m

etric

s)if

link

cost

s cha

nge,

that

info

rmat

ion

mus

t con

verg

e of

co

urse

we

typi

cally

onl

y us

e on

e m

etri

c w

ithin

a d

omai

nw

e ty

pica

lly o

nly

use

stat

ic m

etri

cs, n

ot d

ynam

ic

met

rics

(e.

g., n

ot c

onge

stio

n, b

ut h

op c

ount

)w

e co

uld

have

mul

tiple

met

rics i

n us

e?!

–sa

y ho

p co

unt,

pow

er u

se, r

adio

stre

ngth

for m

obile

wire

less

no

des?

ques

tion:

wou

ld m

ore

com

plex

alg

orith

ms (

if po

ssib

le)

that

dyn

amic

ally

acc

ount

for l

ink

cost

s do

qual

itativ

e be

tter j

ob th

an c

urre

nt si

mpl

e al

gorit

hms o

r jus

t use

ba

ndw

idth

?

50Ji

m B

inkl

ey

type

of s

ervi

cem

y pa

cket

s bef

ore

your

pac

kets

!m

ight

wan

t to

prio

ritiz

e ce

rtain

traf

fic c

lass

es;

.e.g

., 1

cont

rol.,

2. i

soch

rono

us.,

3 bu

rsty

polic

y-ba

sed

rout

ing

(pb

cons

train

ts) -

decr

ee a

ce

rtain

pat

h, o

r out

law

a c

erta

in p

ath

–so

urce

and

stat

ic ro

utin

g ca

n be

use

ful h

ere

–B

GP

clai

ms t

o do

“po

licy”

-rec

ogni

ze th

at m

any

rout

ing

theo

ry ty

pes t

hink

NO

T EN

UF!

!!

51Ji

m B

inkl

ey

pack

et c

olor

ove

rvie

w1.

3 m

echa

nism

s nee

ded

for p

acke

t cl

assi

ficat

ion

2.so

me

way

to se

tup

a lo

gica

l circ

uit

1.en

d to

end

pro

toco

l or r

oute

r-ro

uter

pro

toco

l1.

pack

ets w

ith c

lass

ifier

X sh

ould

be

treat

ed li

ke

first

-cla

ss p

asse

nger

s2.

per p

acke

t cla

ssifi

er (f

low

ID, m

acad

dres

s)3.

rout

er/s

witc

h sc

hedu

ling

algo

rithm

1.if

pack

et c

lass

ifier

, the

n gi

ve it

a b

uffe

r

52Ji

m B

inkl

ey

issu

es -

som

e m

isc.

one

slo

ad-b

alan

cing

-if w

e ha

ve tw

o eq

ual r

oute

s to

X, c

an w

e sp

lit th

e lo

ad b

etw

een

them

?–

equa

l cos

t mul

ti pa

thm

igra

ting

rout

ing

algo

rithm

s -yo

u ha

ve R

IP a

nd n

ow y

ou

wan

t to

switc

h to

OSP

F (a

nd n

o re

boot

ing

mul

tiple

tim

es)

–ca

n yo

u ru

n bo

th?,

sw

itchi

ng o

ne b

y on

e is

dis

rupt

ive

rout

e re

dist

ribu

tion

-at r

outin

g do

mai

n bo

unda

ry, h

ow

exac

tly d

o yo

u ta

ke in

fo fr

om o

ne ro

utin

g pr

otoc

ol a

nd

inje

ct in

to a

noth

er?

(Cis

co ro

ute

map

s)ro

utin

g pa

rtiti

onre

pair

-if t

wo

path

s to

one

net,

and

one

goes

dow

n, c

an ro

uter

s fix

it in

face

of h

iera

rchy

?

53Ji

m B

inkl

ey

a bi

t mor

e on

that

part

ition

–ba

sica

lly m

eans

2 n

ets,

1 lin

k, th

e lin

k bl

ew u

p (o

r ca

me

back

), n

ow w

hat?

»is

ther

e a

way

to g

lue

the

two

nets

toge

ther

thru

som

e ot

her

path

is th

ere

a w

ay to

opt

imiz

e tra

ffic

whe

n th

e ne

ts re

conn

ect i

n or

der t

o m

inim

ize

cont

rol t

raff

ic

aggr

egat

ion

-in

gene

ral,

if ro

utin

g is

con

trol,

can

we

min

imiz

e it

–by

lum

ping

add

ress

es to

geth

er ..

. –

dow

nsid

e./s

: too

muc

h ro

utin

g, to

o m

any

rout

es

54Ji

m B

inkl

ey

type

s: v

ecto

r-di

stan

ce

vect

or-d

ista

nce

algo

rithm

s:“t

ell t

he n

eigh

bors

abo

ut th

e w

orld

”ve

ctor

is d

estin

atio

n (n

et/h

ost)

dist

ance

is m

etric

(hop

coun

t)if

we

calle

d it

dest

inat

ion-

met

ric, o

ther

peo

ple

wou

ld u

nder

stan

d (g

ood

poin

t, D

r. Pe

rlman

)yo

u flo

od y

our d

estin

atio

n, h

opco

unti

nfo

to y

our

dire

ctly

con

nect

ed n

eigh

bor r

oute

rsR

IP is

an

exam

ple,

but

so is

BG

P, I

GR

P, E

IGR

P,

and

rem

embe

r the

fuzz

balls

... (t

ime

as m

etric

)

55Ji

m B

inkl

ey

type

s: li

nk-s

tate

link-

stat

e or

shor

test

pat

h fir

st (S

PF)

“tel

l the

wor

ld a

bout

you

r ne

ighb

ors”

find

out w

ho is

up

loca

lly, a

nd fl

ood

that

in

form

atio

n to

the

entir

e se

t of r

oute

rsth

ey c

an u

se th

e “l

ink-

stat

e”to

bui

ld a

sh

orte

st p

ath

map

to e

very

body

LS is

com

pute

-inte

nsiv

e. V

D is

ban

dwid

th

inte

nsiv

e.

56Ji

m B

inkl

ey

vect

or-d

ista

nce

algo

rithm

exam

ples

: R

IP,

BG

P, IG

RP,

EIG

RP

algo

rithm

ic e

lem

ents

:–

send

: eve

ry N

seco

nds o

ut a

ll co

nnec

ted

inte

rfac

es

broa

dcas

t 2-tu

ples

:(to

net

wor

k X

, ho

p co

unt Y

) ...

–re

cv: i

f new

tupl

e, a

dd to

rout

ing

tabl

eif

bette

r tup

le, c

hang

e ex

istin

gif

“de

ad”

tupl

e, re

mov

e–

timeo

ut: i

f no

refr

esh,

tim

eout

ent

ry in

N *

Y se

cond

broa

dcas

t may

be

lost

, the

refo

re ti

meo

ut is

slow

er

57Ji

m B

inkl

ey

vect

or-d

ista

nce

assu

me

3 ro

uter

s, an

d th

at d

irect

ly

conn

ecte

d ne

ts a

re in

rout

ing

tabl

es to

star

t w

ith.

How

doe

s fol

low

ing

conv

erge

?

r1

r

2

r3

n1

n2

n

3

n4

r1 ta

ble:

(n1

, 1)

(n2,

1)

58Ji

m B

inkl

ey

slow

con

verg

ence

/cou

nt to

infin

ity

vect

or-d

ista

nce

like

this

(RIP

) has

def

ects

chan

ges c

an b

e se

nt w

hen

they

occ

ur, b

ut

mus

t rec

ompu

tea

bit s

o co

nver

genc

e ta

kes

time

(mad

e w

orse

by

poss

ible

loop

s)

coun

t to

infin

ity p

robl

em c

an o

ccur

too

-ro

utin

g lo

op u

ntil

hopc

ount

reac

hes

impo

ssib

le v

alue

59Ji

m B

inkl

ey

coun

t to

infin

ityA

BC

C c

rash

es,

B k

now

s C c

rash

ed b

ut h

asn’

t tol

d A

,bu

t unf

ortu

nate

ly A

talk

s to

B fi

rst

B is

told

by

A:

I can

get

to C

in tw

o ho

ps (a

nd n

ote

it do

esn’

t men

tion

to B

that

the

path

isth

ruB

)B

says

AH

A!,

that

mea

ns I

can

get t

o C

in th

ree

hops

and

repo

rts th

at to

AA

says

AH

A!,

it’s n

ow fo

ur h

ops t

o B

and

tells

Bet

c...

RIP

max

hop

cou

nt (i

nfin

ity) i

s 16

60Ji

m B

inkl

ey

split

-hor

izon

fixu

p(v

ecto

r-di

stan

ce)

A te

lls B

that

its d

ista

nce

to C

is in

finity

–(b

ecau

se B

is th

e di

rect

ion

A g

ets t

he in

fo

from

)w

hen

link

goes

aw

ay, B

will

kno

w th

at

ther

e is

no

path

to C

, and

tell

Ado

esn’

t wor

k in

all

case

s

61Ji

m B

inkl

ey

link-

stat

e al

gorit

hm“t

ell t

he w

orld

abo

ut y

our n

eigh

bors

”ex

ampl

es: O

SPF,

IS-

IS,

NLS

P, I

DPR

link-

stat

e re

quire

s eac

h pa

rtici

patin

g ro

uter

to

keep

map

of c

ompl

ete

topo

logy

in 3

par

ts–

1. d

eter

min

e ne

ighb

or c

onne

ctiv

ity–

2. se

nd (“

flood

”) li

nk-s

tate

pac

ket t

hat s

tate

s whi

ch li

nk

neig

hbor

s are

up

–3.

use

Dijk

stra

shor

test

-pat

h fir

st to

com

pute

bes

t pat

h to

that

net

wor

k

62Ji

m B

inkl

ey

link-

stat

e#de

term

ine

link-

stat

e

“pin

g”ne

ighb

ors t

o de

term

ine

if th

ey a

re u

p or

they

may

bro

adca

st (m

ultic

ast)

thei

r ex

ista

nce

“i m

ultic

ast,

ther

efor

e i a

m...

”C

AB

63Ji

m B

inkl

ey

link-

stat

e#se

nd L

SP

each

par

ticip

atin

g ro

uter

“flo

ods”

(ver

y ca

refu

lly) r

outin

g do

mai

n w

ith L

SPls

p

64Ji

m B

inkl

ey

link-

stat

e#co

mpu

te sh

orte

st p

ath

each

par

ticip

atin

g ro

uter

take

s LSP

s, st

ores

th

em, a

nd c

ompu

tes s

horte

st p

ath

to se

nder

src

dest

65Ji

m B

inkl

ey

link-

stat

e: p

ros/

cons

pros –co

nver

ges f

aste

r, no

cou

nt to

infin

ity p

robl

em +

rout

er

can

forw

ard

LSP

imm

edia

tely

, mus

t rec

ompu

teD

V–

mor

e fu

nctio

nalit

y; e

.g.,

each

rou

ter

has m

ap o

f net

, ca

n m

ake

netw

ork

debu

ggin

g ea

sier

cons

–m

ore

com

pute

than

vd

(doe

s thi

s mat

ter?

)

toss

ups

–ba

ndw

idth

? v

dbr

oadc

asts

sum

mar

y ve

rsio

n of

rout

e ta

ble,

lsr

oute

rs se

nd L

SP a

roun

d ne

t

66Ji

m B

inkl

ey

basi

c to

ols f

or a

dmin

s:pi

ng, b

asic

reac

habi

lity,

pkt

loss

, lat

ency

trac

erou

te, l

aten

cy a

nd h

op c

ount

–re

mem

ber p

aths

may

be

asym

met

ric–

trace

rout

e “-

g”m

ay/m

ay n

ot w

ork

ttcp

, mem

ory

to m

emor

y xf

ersp

eed

–C

isco

is b

uild

ing

it in

to IO

S no

w–

can

mea

sure

tcp-

base

d th

rupu

t, al

so U

DP

pack

et fl

ing

(o

ne w

ay)..

sniff

ers,

free

tcpd

ump

(ww

w.tc

pdum

p.or

g).

67Ji

m B

inkl

ey

high

er-le

vel w

onde

rsve

ry w

ell m

ay in

volv

e SN

MP

–M

RT

Gan

d its

des

cend

ants

»C

ricke

t, a

nd o

ther

fron

t-end

s

–gr

aphs

SN

MP

inte

gers

on

strip

-cha

rts»

port

byte

cou

nts i

n/ou

t

Com

mer

cial

tool

s–

com

mer

cial

mon

itors

, sni

ffer

s, in

trusi

on d

etec

tion

–sn

mp

RM

ON

pro

bes o

r rou

gh/fr

ee e

quiv

alen

ts (n

top)

–ne

twor

k m

gmt.

GU

I won

ders

(HP

open

view

/Cis

co/IB

M, N

etw

ork

Ass

ocia

tes,

etc.

, etc

.)