Chapter3
Transcript of Chapter3
1
Connection-Oriented Networks - Harry Perros 1
Chapter 3:ATM Networks
TOPICS– The ATM header– The ATM protocol stack– The physical layer– ATM switch architectures– ATM adaptation layers– IP over ATM
Connection-Oriented Networks - Harry Perros 2
Asynchronous Transfer Mode (ATM)
• The word Asynchronous in ATM is incontrast to Synchronous Transfer Mode(STM) that was proposed earlier on, whichwas based on the SONET/SDH hierarchy.
• Transfer Mode refers to a telecommunicationtechnique
2
Connection-Oriented Networks - Harry Perros 3
• ATM was standardized by ITU-T (old CCITT) in1988 as the transfer mode of B-ISDN
• It can carry a variety of different types of traffic,such as– Voice– Video– Data
At speeds varying from fractional T1 to 2.4 Gbps
Connection-Oriented Networks - Harry Perros 4
• These different types of traffic havedifferent Quality-of-Service (QoS)requirements, such as:– Packet loss– End-to-end delay
• ATM, unlike IP networks, can provide eachtraffic connection a different type of qualityof service.
3
Connection-Oriented Networks - Harry Perros 5
Some features of ATM
• Connection-oriented packet-switched network• Fixed cell (packet) size of 48+5 bytes
• No error protection on a link-by-link• No flow control on a link-by-link• Delivers cells in the order in which they were
transmitted
Header Payload
5 bytes 48 bytes
Connection-Oriented Networks - Harry Perros 6
The structure of the ATM cell
GFC VPI
VPI VCI
VCI
VCI
HEC
PTI CLP
Informationpayload
1 2 3 4 5 6 7 81
2
3
4
5
.
.
.
53
Byte
UNI cell format
VPI
VPI VCI
VCI
VCI
HEC
PTI CLP
Informationpayload
1 2 3 4 5 6 7 81
2
3
4
5
.
.
.
53
Byte
NNI cell format
4
Connection-Oriented Networks - Harry Perros 7
Fields in the ATM cell header
• GFC• Connection identifier: VPI/VCI,• Payload type indicator (PTI)• Cell loss priority (CLP)• Head error control (HEC)
Connection-Oriented Networks - Harry Perros 8
ATM connections
• Identified by the combined fields– virtual path identification (VPI), and– virtual channel identification (VCI)
• VPI field:– 256 virtual paths at the UNI interface, and – 4096 virtual paths at the NNI interface.
• VCI field:– a maximum of 65,536 VCIs.
5
Connection-Oriented Networks - Harry Perros 9
• VPI/VCI values have local significance.That is, they are only valid for a single hop.
• A connection over many hops, is associatedwith a different VPI/VCI value on each hop.
• Each switch maintains a switching table.For each connection, it keeps the incomingand outgoing VPI/VCI values and the inputand output ports.
Connection-Oriented Networks - Harry Perros 10
Label swapping
VPI=10VCI=89
VPI=30VCI=53
VPI=100VCI=53
VPI=50VCI=77
VPI=30VCI=41
30 53 4 100 53 5
30 41 1 30 53 440 62 2 10 89 3
10 89 1 50 77 6
ATMswitch 1
ATMswitch 2
ATMswitch 3
VPI=40VCI=62
A
B
C
D2
1
3
4 4
1
5
6
6
Connection-Oriented Networks - Harry Perros 11
PVCs and SVCs• Depending how a connection is set-up, it
may be– Permanent virtual circuit (PVC)– Switched Virtual circuit (SVC)
• PVCs are set-up administratively. Theyremain up for a long time.
• SVCs are set-up in real-time using ATMsignalling. Their duration is arbitrary.
Connection-Oriented Networks - Harry Perros 12
Payload type Indicator
• PTI Meaning • 000 User data cell, congestion not experienced, SDU type=0• 001 User data cell, congestion not experienced, SDU type=1• 010 User data cell, congestion experienced, SDU type=0• 011 User data cell, congestion experienced, SDU type=1• 100 Segment OAM flow-related cell• 101 End-to-end OAM flow-related cell• 110 RM cell• 111 Reserved
7
Connection-Oriented Networks - Harry Perros 13
No error detected(No action)
Error detected(cell discard)
Head Error Control (HEC)
Correctionmode
DetectionmodeNo error detected
Single bit error detected(correction)
Multiple bit error detected(cell discarded
Connection-Oriented Networks - Harry Perros 14
The ATM protocol stack
ATM adaptation layer
ATM layer
Physical layer
voice Video Data
8
Connection-Oriented Networks - Harry Perros 15
The physical layer
• The physical layer transports ATM cellsbetween two adjacent ATM layers.
• It is subdivided into– transmission convergence (TC) sublayer– physical medium-dependent (PMD) sublayer.
Connection-Oriented Networks - Harry Perros 16
The transmission convergence (TC)sublayer
• HEC cell generation and verification– Implements the HEC state machine
• Decoupling of cell rate– Maintains a continuous bit stream by inserting idle cells
• Transmission frame generation and recovery– Such as SONET frames
• Cell delineation
9
Connection-Oriented Networks - Harry Perros 17
Cell delineation is the extraction of cells from thebit stream received from the PMD sublayer.
hunt
Presync
Sync
CorrectHEC
CorrectHEC for δ cells
IncorrectHEC for α cells
IncorrectHEC
Connection-Oriented Networks - Harry Perros 18
• Physical medium dependent (PMD)– Timing function
• Used to synchronize the transmitting and receivingPMD sublayers.
– Encoding/decoding• PMD may operate on a bit-by-bit basis or using
block coding such as 4B/5B and 8B/10B schemes.
10
Connection-Oriented Networks - Harry Perros 19
ATM physical layer interfaces
• SONET/SDH• Plesiochronous digital hierarchy (PDH)• Nx64 Kbps• Inverse mulitplexing for ATM (IMA)• asymmetric digital subscriber line (ADSL)• APON
Connection-Oriented Networks - Harry Perros 20
The ATM layer
• The ATM layer is concerned with the end-to-end transfer of information, i.e., from thetransmitting end-device to the receivingend-device.
• Below, we summarize its main features.
11
Connection-Oriented Networks - Harry Perros 21
Connection-oriented packet switching
• The ATM layer is a connection-oriented point-topoint packet-switched network with fixed-sizepackets (known as cells).
• A connection is identified by a series of VPI/VCIlabels, as explained above, and it may be point-to-point or point-to-multipoint.
• Cells are delivered to the destination in the orderin which they were transmitted.
Connection-Oriented Networks - Harry Perros 22
Cell switching in ATM networksis carried out at the ATM level
End-device
Application Application
ATM switch End-deviceATM switch
12
Connection-Oriented Networks - Harry Perros 23
No error and flow controlon each hop
• Low probability of a cell getting lost or delivered to thedestination end-device in error.
• The recovery of the data carried by lost or corrupted cellsis expected to be carried out by a higher-level protocol,such as TCP.
• When TCP/IP runs over ATM, the loss or corruption of thepayload of a single cell results in the retransmission of anentire TCP PDU.
Connection-Oriented Networks - Harry Perros 24
Addressing
• Each ATM end-device and ATM switch has a uniqueATM address.
• Private and public networks use different ATMaddresses. Public networks use E.164 addresses andprivate networks use the OSI NSAP format.
• ATM addresses are different to IP addresses.
13
Connection-Oriented Networks - Harry Perros 25
Quality of service
• Each ATM connection is associated with aquality-of-service category.
• Each quality-of-service category is associated witha set of traffic parameters and a set of quality-of-service parameters.
• The ATM network guarantees the negotiatedquality-of-service for each connection.
Connection-Oriented Networks - Harry Perros 26
Congestion control
• In ATM networks, congestion controlpermits the network operator to carry asmuch traffic as possible without affectingthe quality of service requested by the users.
• It consists of call admission control and apolicing mechanism.
14
Connection-Oriented Networks - Harry Perros 27
A generic ATM switch
Incominglinks
. . .
. . . Outgoinglinks
Inputqueues
Outputqueues
CPU
Switch fabric
1
N N
1
The ATM switch architecture
Connection-Oriented Networks - Harry Perros 28
The shared memory ATM switcharchitecture
. . .
. . .
Shared memory1
N N
1
15
Connection-Oriented Networks - Harry Perros 29
• If the speed of transmission on eachincoming and outgoing link is V, then theswitch can keep up at maximum arrival rate,if the memory's bandwidth is 2NV
• Total memory capacity is B cells• Each linked list i is associated with a
minimum dedicated space and it is limitedto an upper bound Bi, Bi<B so that ΣBi>B.
Connection-Oriented Networks - Harry Perros 30
Cell loss in a shared memory switch
• Cell loss occurs when a cell arrives at a time whenthe shared memory is full, that is it, contains Bcells.
• Cell loss can also occur when a cell withdestination output port i arrives at a time when thetotal number of cells queued for this output port isBi cells (even if the total number of cells in theshared memory is less than B.)
16
Connection-Oriented Networks - Harry Perros 31
Non-blocking output buffering switch• In a non-blocking switch, the switching fabric
does not give rise to internal or external blocking.• An output buffering switch has buffers only at its
output ports.
. . .
. . .
output ports
• A shared memory switch is non-blocking withoutput buffering
Connection-Oriented Networks - Harry Perros 32
Scheduling algorithms
• Let us consider a non-blocking switch with outputbuffering. Each output buffer holds cells thatbelong to different connections.
• Each of these connections is associated with aquality-of-service category.
• The cells belonging to these connections aregrouped into queues, one per quality-of-servicecategory, and these queues are served using ascheduling algorithm.
17
Connection-Oriented Networks - Harry Perros 33
Static priorities
From switchfabric
CBR
RT-VBR
NRT-VBR
ABR
UBR
Output port
• Priorities among thequeues
• Always serve highpriority queue first,then second priorityqueue, etc.
• Aging• Purging
Connection-Oriented Networks - Harry Perros 34
Early deadline first (EDF) algorithm
• Each cell is assigned a deadline upon arrival atthe buffer. The scheduler servers the cellsaccording to their deadlines, so that the one withthe earliest deadline gets served first.
• Cells belonging to delay-sensitive applications,such as voice or video, can be served first byassigning them deadlines closer to their arrivaltimes.
18
Connection-Oriented Networks - Harry Perros 35
The round-robin scheduler
• Each output buffer is organized into a number oflogical queues.
• The scheduler serves one cell from each queue ina round robin fashion
• Empty queues are skipped• Weighted round robin scheduling can be used to
serve a different number of cells from each queue
Connection-Oriented Networks - Harry Perros 36
The ATM adaptation layer
• The purpose of AAL is to isolate higherlayers from the specific characteristics ofthe ATM layer.
• AAL consists of the– convergence sublayer, and the– segmentation-and-reassembly sublayer.
19
Connection-Oriented Networks - Harry Perros 37
The AAL sublayers
SAP
Service SpecificConvergence Sublayer (SSCS)
Common Part Sublayer (CPS)
Segmentation and Reassembly
Convergence Sublayer
SAP
Connection-Oriented Networks - Harry Perros 38
ATM Adaptation Layer 1 (AAL 1)
• This AAL can be used for applications such as:– circuit emulation services
• It emulates a point-to-point TDM circuit over ATM– Constant-bit rate audio
• Used to provide an interconnection between twoPBXs over a private or public ATM network
20
Connection-Oriented Networks - Harry Perros 39
The SAR encapsulation for AAL 1
payloadSAR Header
47 bytesSN SNP
CRC-3
3 bits
Parity
1 bit3 bits1 bit
CSI Sequence. count
Connection-Oriented Networks - Harry Perros 40
The AAL 1 CS functions:1. Handling of cell variation
– Due to queueing delays, inter-arrival times of cells vary (jitter).
– CS writes received data into a buffer, and then delivers theinformation to the application at constant bit rate.
ATM cloud
Inter-departure gaps
Sender
cell i-1
Inter-arrival gaps
Receiver
cell i
celli+1
cell i-1
cell i
celli+1
t i-1 t i si-1 si
21
Connection-Oriented Networks - Harry Perros 41
2. Processing of the sequence count– The sequence count values are processed by CS in
order to detect lost or misinserted cells. Detectedmisinserted cells are discarded. In order to maintainbit count integrity of the AAL user information, itmay be necessary to compensate for lost cells byinserting dummy SAR-PDU payloads.
3. Forward error correction– For video and high quality audio forward error
correction may be performed in order to protectagainst bit errors. This may be combined withinterleaving of AAL user bits to give a more secureprotection against errors.
Connection-Oriented Networks - Harry Perros 42
4. Transfer of timing informationa. Synchronous residual time stamp (SRTS):
CS conveys to the receiver in the CSI field thedifference between a common clock derived from thenetwork and the sender’s clock
b. Adaptive clock method:The receiver writes the received information into abuffer and reads out from the buffer. If its clock isfast/slow the occupancy in the buffer will bebelow/over the median
22
Connection-Oriented Networks - Harry Perros 43
5. Structured and unstructured data transfersTwo CS-PDU formats have been defined:a.CS-PDU non-P format:
Constructed from 47 bytes of informationsupplied by an AAL user
b. CS-PDU P format:Constructed from a 1-byte header and 46 bytesof information supplied by an AAL use.The header consists of a 7-bit pointer (SDTpointer) and 1 even bit parity.
Connection-Oriented Networks - Harry Perros 44
Circuit Emulation Services
IWFA
IWFBATM network
UNI UNI
CBRUser A
CBRUser B
• The structured and unstructured data transfers areused in Circuit Emulation Services (CES), whichemulate a T1/E1 link over ATM.
• CES is implemented in an interworking function(IWF).
23
Connection-Oriented Networks - Harry Perros 45
• Unstructured service– The entire DS1/E1 signal is transported by
packing it bit by bit into the 47-byte payload ofa CS-PDU non-P format, which is then carriedby an ATM cell.
47 bytes -> 376 bits -> less than 2 DS-1 frames (193 bits/frame)
Connection-Oriented Networks - Harry Perros 46
• Structured transfers– It is used to carry fractional T1/E1, i.e. Nx64
Kbps– Fractional T1/E1 generates blocks of N bytes
every 125 µsec. Such a block of data is referredto in the standards as a structured block.
– Blocks of N bytes are transported back-to-backover successive cells using both the CS-PDUnon-P and P formats.
24
Connection-Oriented Networks - Harry Perros 47
• The SDT pointer– The SDT pointer in the CS-PDU P format is
used to help delineate the boundaries of theseblocks.
– The actual rules as to when to use the SDTpointer in the P format are somewhat complex.
Connection-Oriented Networks - Harry Perros 48
An example: block size = 20 bytes
Seq, count 0 Seq, count 1 Seq, count 2 Seq, count 3
Seq, count 4 Seq, count 5 Seq, count 6 Seq, count 7
Seq, count 0 Seq, count 1 Seq, count 2 Seq, count 3
20 20 6 14 20 13 7 20 20 20 20 7
13 20 14 6 20 20 19 20 7 13 20 14
6 20 20 20 20 7 13 20 14 6 20 20 1
1
25
Connection-Oriented Networks - Harry Perros 49
ATM Adaptation Layer 2 (AAL 2)
• Defined for delay sensitive applicationswith a low bit rate, such as voice andvoiceband traffic (facsimile, modem traffic)
• AAL 2 is used in to interconnect two distantpublic or private telephone networks overan ATM network.
Connection-Oriented Networks - Harry Perros 50
• At the sender, AAL 2 multiplexes several streamsonto the same ATM connection
• At the receiver, it de-multiplexes the date fromthe connection to the individual streams.
26
Connection-Oriented Networks - Harry Perros 51
The SSCS and CPS sublayers
• The AAL 2 services are provided by theconvergence sublayer, which is subdivided intothe– Service Specific Convergence Sublayer (SSCS)– Common part sublayer (CPS).
Connection-Oriented Networks - Harry Perros 52
Functional model of AAL 2 (sender side)
• Each stream is served by a separate SSCS which isassociated with a CID
SSCS
AAL-SAP
ATM-SAP
SSCSSSCS
CID=XCID=Y
CID=Z
CPS
27
Connection-Oriented Networks - Harry Perros 53
SSCS for AAL 2 trunking
• A specialized SSCS has been developed tosupport “ATM trunking using AAL 2 fornarrowband services”.
• It is described in Chapter 12
Connection-Oriented Networks - Harry Perros 54
• A transmitting SSCS uses a timer to decidewhen to pass on the data to CPS.
• Data from an SSCS is packed into a CPS-packet• CPS-packets from different SSCSs are packed
into a CPS-PDU, which is exactly 48 bytes andit is carried in an ATM cell
CPS-packets and CPS-PDUs
28
Connection-Oriented Networks - Harry Perros 55
Packing CPS-packets into CPS-PDUs
CPS-PDUs
1 2 3 4 5
54321 3
ATM cells
CPS-packets
Connection-Oriented Networks - Harry Perros 56
The structure of the CPS-packetand CPS-PDU
OSFP
CPS-PDUpayload
SN
PAD
C I D
PPT LI
HEC UUI
CPS-packetpayload
1
2
3
1
48
CPS-packet CPS-PDU
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
29
Connection-Oriented Networks - Harry Perros 57
The CPS-packet fields• Channel identifier (CID) - 8 bits:
– Identifies a channel. Same value is used for both directions.– CIDs are allocated using the AAL negotiation procedures (ANP)
• Packet payload type (PPT) - 2 bits:– Indicates whether it carries voice or network management data
• Length indicator (LI) - 6 bits:– Default maximum length of the CPS-packet payload is 45 bytes.
• Header error control (HEC) - 5 bits:– Pattern is: x5+x2+1.
• User-to-user-indication (UUI) - 3 bits:– Used to transfer information transparently between the peers.
Connection-Oriented Networks - Harry Perros 58
CPS-PDU fields
Offset field (OSF) (6 bits)• Used to identify the beginning of a CPS-
packet. It points to the first new CPS-packetin the CPS-PDU payload
• In the absence of a new CPS-packet, itpoints to the beginning of the pad
• The value of 47 indicates that there is nobeginning of a CPS-packet in the CPS-PDU.
30
Connection-Oriented Networks - Harry Perros 59
An example
20 48 35 20
OSF =0 OSF=21 OSF=9
20 27 21 26 9 20 #1#2
#2 #3 #3 #4padding
Connection-Oriented Networks - Harry Perros 60
ATM adaptation layer 5 (AAL 5)
• Very popular AAL due to its simplicity• A user-PDU is encapsulated and then broken up to
fragments, each carried by an ATM cell• AAL 5 consists of
– Convergence sublayer (CS)• SSCS• CPS
– Segmentation and reassembly (SAR).
31
Connection-Oriented Networks - Harry Perros 61
• Provides a non-assured transfer operation.• User-PDUs of a length up to 65,535 bytes
can be transferred.• Erroneous CPS-PDUs can be detected at the
receiver’s side. No recovery of an erroneousCS-PDU takes place. Instead, an indicationis sent to the the higher-level application.
CPS
Connection-Oriented Networks - Harry Perros 62
• Pad: from 0 to 47 bytes, so that the entire CPS-PDU becomesan integer multiple of 48 bytes. The User-PDU can be up to65,535 bytes
• CPS User-to-user indication (CPS-UU): 1-byte field• Common part indicator (CPI): 1-byte field for future use• Length: 2-byte field gives length of CPS-PDU payload.• CRC pattern: 4-byte field contains the FCS calculated using
the pattern x32+x26+x23+x22+x16+x12+x11+x10+x8+x7+x5+x4+x2+x+1.
User-PDU Pad CRC-32LengthCPS-UU
CPS encapsulation
32
Connection-Oriented Networks - Harry Perros 63
SAR sublayer - transmitter
• SAR segments a CPS-PDU into a sequenceof 48-byte segments.
• No additional encapsulation• Each segment is carried in the payload of an
ATM cell• Last cell has its PTI marked with SDU=1.
Connection-Oriented Networks - Harry Perros 64
SAR sublayer - Receiver
• SAR appends payloads of the ATM cellinto a buffer until1. It encounters an SDU=1 in PTI field
• It checks the CRC and then passes the PDU to theapplication above with an indication as to whetherit is correct or not.
2. Buffer is exceeded• It passed the PDU to the application above with an
indication that buffer was exceeded.
33
Connection-Oriented Networks - Harry Perros 65
Classical IP and ARP over ATM
• This is a technique proposed by IETF forsupporting IP over ATM in a single logical IPsubnet (LIS).
• A LIS is a group of IP hosts that share a commonIP network number and subnet mask, and they allcommunicate with each other directly over ATMconnections.
Connection-Oriented Networks - Harry Perros 66
Group of host with the same network addressand same subnet maskIP address: 193.14.0.0
We now replace the transport network with ATM switches
A logical IP subnet (LIS)
34
Connection-Oriented Networks - Harry Perros 67
PHYATMAALCIPIP
TCP
PHYATMAALCIPIP
TCP
IP packet
CS-PDU
SAR
ATM
IP packet
CS-PDU
SAR
ATM
Computer Computer
Classical IP
ATM switch
Connection-Oriented Networks - Harry Perros 68
ATMARP and InATMARP
• A LIS member has both an IP and an ATMaddress.
• IP addresses are resolved to ATM addresses usingthe ATMARP protocol within the LIS (based onARP).
• The inverse ATMARP (InATMARP) protocol isused to resolve an ATM address to an IP address.(It is similar to the inverseARP).
35
Connection-Oriented Networks - Harry Perros 69
The ATMARP server
• It maintains a table or a cache of IP and ATMaddress mappings
• It learns about the IP and ATM addresses of theLIS members (IP clients) through the messagesexchanged between ATMARP and the LIS hosts.
• It typically resides on an ATM switch. (An ATMswitch load commonly contains the ATMARPserver as well).
Connection-Oriented Networks - Harry Perros 70
ATMARP messages
• ATMARP_request: Used to request the ATMARP serverthe ATM address of a destination IP client.
• ATMARP_reply: Used by the ATMARP server to respondto an IP client with the destination ATM address.
• InATMARP_request: Sent from the ATMARP server to anIP client to obtain its IP address.
• InATMARP_reply: This is the response from an IP clientwith its IP address.
• ATMARP_NAK: Negative response issued by theATMARP server to a requesting IP client.
36
Connection-Oriented Networks - Harry Perros 71
Registration
• An IP client must first register its IP and ATMaddresses with the ATMARP server.– The client establishes a connection to the ATMARP
server (it knows the ATMARP server’s ATM address).– It then transmits an ATMARP_request, asking its own
ATM address.– The ATMARP server checks against duplicate entries
in its table, time stamps the entry, adds it to its table,and sends an ATMARP_reply. The entry is valid for aminimum of 20 minutes and has to be refreshed.
Connection-Oriented Networks - Harry Perros 72
Address resolution• An ATMARP client 1 can communicate with an
ATMARP client 2 immediately if it knows itsATM address.
• If the destination ATM address is not known,client 1 invokes the ATMARP process.– It sends an ATMARP_request to the ATMARP server.– If the server has the requested address in its table, it
returns an ATMARP_reply.– Otherwise, it returns an ATMARP_NAK.