ATM* ATM (Asynchronous Transf er Mode) is t he swit ching and t - - PDF document

ATM*

ATM (Asynchronous Transf er Mode) is t he

swit ching and t ransport t echnology of t he B-I SDN (Broadband I SDN) archit ect ure (1980)

Goals: high speed access t o business and

resident ial users (155Mbps t o 622 Mbps); integrated services support (voice, dat a, video, image)

* Kurose and Ross, “Computer Networking”

ATM VCs

Focus on bandwidth allocation f acilit ies (in

cont rast t o I P best ef f ort )

ATM main role t oday: “switched” link layer f or I P-

• ver- ATM

ATM is a virtual circuit t ransport : cells (53 byt es)

ar e car r ied on VCs

in I P over ATM: Permanent VCs (PVCs) bet ween I P

rout ers;

scalabilit y problem: N(N-1) VCs bet ween all I P

rout er pairs

ATM VCs

Switched VCs (SVCs) used f or short lived

connect ions

Pros of ATM VC approach:

• Can guarant ee QoS perf ormance t o a connect ion mapped

t o a VC (bandwidt h, delay, delay j it t er) Cons of ATM VC approach:

• I nef f icient support of datagram t raf f ic; P

VC solut ion (one P VC bet ween each host pair) does not scale;

• SVC int roduces excessive latency on short lived

connect ions

• High SVC processing Overhead

ATM Address Mapping

Rout er int erf ace (t o ATM link) has two addresses:

I P and ATM address.

To rout e an I P packet t hrough t he ATM net work,

t he I P node:

(a) inspect s own rout ing t ables t o f ind next I P router address (b) t hen, using ATM ARP table, f inds ATM addr of next rout er (c) passes packet (wit h ATM address) t o ATM layer At t his point , t he ATM layer t akes over: (1) it det ermines t he interf ace and VC on which t o send out t he packet (2) if no VC exist s (t o t hat ATM addr) a SVC is set up

ATM P hysical Layer

Two Physical sublayers: (a) Physical Medium Dependent (PMD) sublayer

• (a.1) SONET/ SDH: t ransmission f rame st ruct ure (like a

cont ainer carrying bit s);

• bit synchronizat ion;
• bandwidt h part it ions (TDM);
• sever al speeds: OC1 = 51.84 Mbps; OC3 = 155.52 Mbps;

OC12 = 622.08 Mbps

• (a.2) TI / T3: t ransmission f rame st ruct ure (old

t elephone hierarchy): 1.5 Mbps/ 45 Mbps

• (a.3) unstructured: j ust cells (busy/ idle)

ATM P hysical Layer (more)

Second physical sublayer

(b) Transmission Convergence Sublayer (TCS): it adapt s PMD sublayer t o ATM t ransport layer

TCS Funct ions:

• Header checksum generat ion: 8 bit s CRC; it prot ect s a 4-

byt e header; can correct all single errors.

• Cell delineation
• Wit h “unst ruct ured” P

MD sublayer, t ransmission of idle cellswhen no dat a cells are available in t he t ransmit queue

ATM Layer

ATM layer in charge of t ransport ing cells across

t he ATM net work

ATM layer prot ocol def ines ATM cell header

f ormat (5byt es);

payload = 48 byt es; t ot al cell lengt h = 53 byt es

ATM Layer

VCI (virt ual channel I D): t ranslat ed f rom link t o

link;

PT (P

ayload t ype): indicat es t he t ype of payload (eg mngt cell)

CLP (Cell Loss Priorit y) bit : CLP = 1 implies t hat

t he cell is low priorit y cell, can be discarded if rout er is congest ed

HEC (Header Error Checksum ) byt e

ATM Adapt at ion Layer (AAL)

ATM Adaptation Layer (AAL): “adapt s” t he ATM

layer t o t he upper layers (I P or nat ive ATM applicat ions)

AAL is present only in end systems, not in

swit ches

The AAL layer has it s header/ trailer f ields,

carried in t he ATM cell

ATM Adapt ion Layer (AAL) [more]

Dif f erent versions of AAL layers, depending on

t he service t o be support ed by t he ATM t ransport :

• AAL1: f or CBR (Const ant Bit Rat e) services such as

circuit emulat ion

• AAL2: f or VBR (Variable Bit Rat e) services such as MP

EG video

• AAL5: f or dat a (eg, I P dat agrams)

ATM Adapt ion Layer (AAL) [more]

Two sublayers in AAL:

• (Common Part) Convergence Sublayer (cpcs):

encapsulat es I P payload

• Segmentation/ Reassembly Sublayer (sar ):

segment s/ reassembles t he CPCS (of t en quit e large, up t o 65K byt es) int o 48 byt e ATM segment s

AAL5 - Simple And Ef f icient AL (SEAL)

AAL5: low overhead AAL used t o carry I P

dat agrams

• SAR header and t railer eliminat ed; CRC (4 byt es) moved

t o CPCS

• P

AD ensures payload mult iple of 48byt es (LENGTH = P AD byt es)

• At dest inat ion, cells are reassembled based on VCI

number; AAL indicate bit delineat es t he CP CS-P DU; if CRC f ails, P DU is dropped, else, passed t o Convergence Sublayer and t hen I P

Dat agram J ourney in I P-over-ATM Net work

At Source Host:

• (1) I P

layer f inds t he mapping bet ween I P and ATM exit address (using ARP ); t hen, passes t he dat agram t o AAL5

• (2) AAL5 encapsulates dat g and it segments t o cells;

t hen, down t o ATM I n the network, t he ATM layer moves cells f rom

swit ch t o swit ch, along a pre- established VC

At Destination Host, AAL5 reassembles cells

int o original dat g;

• if CRC OK, dat gram is passed up t he I P prot ocol.

ARP in ATM Net s

ATM can rout e cells only if it has t he ATM

addr ess

• Thus, I P

must translate exit I P address t o ATM address The I P/ ATM addr t ranslat ion is done by ARP

(Addr Recogn P rot ocol)

Generally, ATM ARP t able does not st ore all ATM

addresses: it must discover some of t hem

Two techniques:

• br oadcast
• ARP ser ver s

ARP in ATM Net s (more)

(1) Broadcast t he ARP

request t o all destinations:

• (1.a) t he ARP Request msg is broadcast t o all ATM

dest inat ions using a special broadcast VC;

• (1.b) t he ATM dest inat ion which can mat ch t he I P

address ret urns (via unicast VC) t he I P / ATM address map; Broadcast overhead prohibit ive f or large ATM

net s.

ARP in ATM Net s (more)

(2) ARP Server:

• (2.a) source I P router f orwards ARP

request t o server

• n dedicat ed VC (Not e: all such VCs f rom rout ers t o ARP

have same I D)

• (2.b) ARP server responds t o source rout er wit h

I P/ ATM t ranslat ion Host s must register t hemselves wit h t he ARP

server Comments: mor e scaleable t han ABR Broadcast approach (no broadcast st orm). However, it requires an ARP server , which may be swamped wit h request s

X.25 and Frame Relay

Wide Area Net work t echnologies (like ATM); also,

bot h Virtual Circuit orient ed , like ATM

• X. 25 was born in mid ‘70s, wit h t he support of

t heTelecom Carriers, in response t o t he ARP ANET dat agram t echnology (religious war..)

Frame relay emerged f rom I SDN t echnology (in

lat e ‘80s)

Bot h X.25 and Frame Relay can be used t o carry

I P datagrams; t hus, t hey are viewed as Link Layers by t he I P prot ocol layer (and are t hus covered in t his chapt er)

X.25

X.25 builds a VC bet ween source and dest inat ion

f or each user connection

Along t he pat h, error control (wit h

ret ransmissions) on each hop using LAP-B, a variant of t he HDLC prot ocol

Also, on each VC, hop by hop f low control using

credits;

• congest ion arising at an int ermediat e node propagat es t o

source via backpressure

X.25

As a result , packet s are delivered reliably and in

sequence t o dest inat ion; per f low credit cont rol guarant ees f air sharing

Put t ing “intelligence into the network” made

sense in mid 70s (dumb t erminals wit hout TCP)

Today, TCP and pract ically error f ree f ibers f avor

pushing t he “intelligence to the edges”; moreover, gigabit rout ers cannot af f ord t he X.25 processing

• ver head

As a result , X.25 is rapidly becoming extinct

Frame Relay

Designed in lat e ‘80s and widely deployed in t he

‘90s

FR VCs have no error control Flow (rate) control is end to end; much less

processing O/ H t han hop by hop credit based f low cont r ol

Frame Relay (more)

Designed t o interconnect corporat e cust omer

LANs

Each VC is like a “pipe” carrying aggregat e t raf f ic

bet ween t wo rout ers

Corporat e cust omer leases FR service f rom a

public Frame Relay net work (eg, Sprint or ATT)

Alt ernat ive, large cust omer may build Private

Frame Relay net work.

Frame Relay (more)

Frame Relay implement s most ly permanent VCs

(aggregat e f lows)

10 bit VC I D f ield in t he Frame header I f I P runs on t op of FR, t he VC I D corresponding

t o dest inat ion I P address is looked up in t he local VC table

FR swit ch simply discards f rames wit h bad CRC

(TCP ret ransmit s..)

Frame Relay -VC Rat e Cont rol

CI R = Commit t ed I nf ormat ion Rat e, def ined f or

each VC and negot iat ed at VC set up t ime; cust omer pays based on CI R

DE bit = Discard Eligibilit y bit in Frame header

• DE bit = 0: high priority, rat e compliant f rame; t he

net work will t ry t o deliver it at “all cost s”

• DE bit = 1: low priority, “marked” f rame; t he net work

discards it when a link becomes congest ed (ie, t hreshold exceeded)

Frame Relay - CI R & Frame Marking

Access Rate: rat e R of t he access link bet ween

source router (cust omer) and edge FR switch (pr ovider ); 64Kbps < R < 1,544Kbps

Typically, many VCs (one per dest inat ion rout er)

mult iplexed on t he same access t runk; each VC has

• wn CI R

Edge FR swit ch measures t raf f ic rat e f or each

VC; it marks

(ie DE <

= 1) f rames which exceed CI R (t hese may be lat er dr opped)

Frame Relay - Rat e Cont rol

Frame Relay provider “almost ” guarant ees CI R rat e (except

f or overbooking)

No delay guarant ees, even f or high priorit y t raf f ic Delay will in part depend on rat e measurement interval Tc;

t he larger Tc, t he burst ier t he t raf f ic inj ect ed in t he net work, t he higher t he delays

Frame Relay provider must do caref ul traf f ic engineering

bef ore commit t ing t o CI R, so t hat it can back up such commit ment and prevent overbooking

Frame Relay CI R is t he f irst example of t raf f ic rat e

dependent charging model f or a packet swit ched net work