TCP/IP over ATM
LESSON 18

What is ATM ??

• Asynchronous Transfer Mode (ATM):
  • Is a lower-level communications technology which can perform at very high speeds.
• ATM is integration of multiple media types:
  • Data
  • Voice
  • Video
  • Images

Circuit vs. Packet Switching:

• Circuit:
  • End-to-end connection.
  • Constant delay.
  • Information is delivered in order (sequenced delivery).
  • Network is transparent to the user's information.
  • May be inefficient for some types of traffic.
• Virtual Circuit:
• End-to-end connection in a packet switched network.
• Packets delivered in order.
• Network is not transparent to what the user is sending.
• More efficient than circuits for certain applications.

ATM Components:

• ATM components include:
  • Hubs
  • Switches
  • Network Interface Adapter Cards
  • Multiplexers
  • Routers / Gateways
  • Access Nodes
  • Concentrators

ATM Network Interfaces:

• ATM components can be interconnected to support larger networks and to provide different interfaces to the same network.
• Two interfaces exist for ATM network components.
  • UNI (User-to-Network Interface):
    • Spans from customer premises to the network switch (common carrier) or can be used to connect switches within a private network.
  • NNI (Network-to-Network Interface):
    • Connects ATM switches together using slightly modified protocols at higher speeds than the user-to-network interface (UNI).
  

ATM Connection Paradigms:

• Hosts must establish a connection over an ATM network before they can transfer any data.
• To do so, they need to use a virtual circuit connection.
• Two forms exist:
  • Permanent Virtual Circuits (PVC):
    • The virtual circuit is manually setup through provisioning of network resources by the network providers.
    • No connection setup is needed prior to any data transmission.
  • Switched Virtual Circuits (SVC):
    • Functions like a voice telephone call in that it must setup a connection prior to user of the network.
    • Signaling is used throughout the ATM network to allow devices to request circuit setup, take-down, and management.
    • SVCs use Meta-signaling.

Paths, Circuits, and Connection Identifiers:

• A circuit identifier used with UNI is a 24-bit integer divided into two fields.
• It uniquely identifies an open virtual circuit on one end of a connection.
  

• Virtual Circuit Identifier (VCI):
  • 16-bits which identifies the virtual circuit used for this particular connection.
• Virtual Path Identifier (VPI):
  • 8-bits which identifies the particular path the virtual circuit identifier is traveling over.
• Many VCI are usually contained within a single VPI.
• If many circuits follow the same path, the system administrator can give those circuits the same VPI value.
• ATM switches use VPI to efficiently route virtual channels.

ATM Cell Transport:

• ATM uses fixed-size packets of 53-octets called cells to carry all types of data over the network.
• Five octets of each cell are used as a header to carry control information.
  
  

  ATM Cell

  Header 5 octets

  Data Area (Payload) 48 octets

ATM Adaptation Layer:

• ATM transfers cells. Applications that transfer data over ATM do not read or write cells.
• Higher-layer protocols interact with ATM through the ATM Adaptation Layer (AAL).
• AAL performs many functions including:
  • Detection and correction of errors.
  • Multiplexing and Demultiplexing of data units.
  • Lost or corrupted cells.
  

• Usually the AAL is located in firmware on the network interface card.
• When a virtual circuit is created over ATM, both end- points must agree on and use the same AAL protocol.
• Several combinations of AAL protocols, services, and sublayers exist: each providing support for different types of transport services.

ATM Adaptation Layer:
Service Classes

• Service Classes:
  • Class A - Circuit emulation; constant bit rate
  • Class B - Variable bit rate; audio and video
  • Class C - Connection-oriented data transfer; variable bit rate (Frame Relay, TCP/IP)
  • Class D - Connectionless data transfer (SMDS)
  • Class X - Connection-oriented; best-effort delivery; user regulated.

ATM Adaptation Layer:
Protocols

• AAL Protocols:

  AAL1 AAL2 AAL3/4, 5 AAL3/4, 5 AAL3/4, 5 AAL0

  = Class A = Class B = Class C = Class D = Class X = Null

ATM Adaptation Layer:
Sublayers

• AAL Sublayers (lowest to highest):
  • Segmentation and Reassembly Sublayer (SAR):
    • Required for all AAL protocols (except AAL 0), SAR handles segmentation and reassembly of data.
  • Common Part Convergence Sublayer (CPCS):
    • Used only in AAL1, 3/4, 5. Handles cell delay variation, sequencing, & error correction/detection.
  • Service Specific Convergence Sublayer (SSCS)
    • Optional in AAL3/4, 5. Used for specific features.

ATM Adaptation Layer:
AAL-1

• AAL1 accepts and sends data across an ATM network at a fixed bit rate.
• It's used for applications like video because it provides a fixed-rate delivery service.
• This is required to guarantee a constant video stream free of pauses or flickering.

ATM Adaptation Layer:
AAL-0, 2, 3/4

• AAL0 - is NULL defined so there is no protocol used. It is a raw interface.
• AAL2 - is not yet complete. The standard definition has not been finished by ITU-T or the ATM Forum.
• AAL3/4 - is a complex protocol which provides a similar service to AAL5. It is not as popular because of this complexity.

ATM Adaptation Layer:
AAL-5

• AAL5 - is used to send conventional data packets across an ATM network.
• It provides an interface that accepts and delivers large, variable-length packets (1..65,535 octets).
• AAL5 packets are divided into ATM cells in exact multiples of 48 octets (including 8-bit trailer).
• If all cells are not completely full, the last one can contain from 0..40 octets of data, padded with zeros to fill it.
  

• AAL5 places control information in the 8 octet trailer, which includes:
  • Length field (16-bits).
  • Cyclic Redundancy Check (32-bit CRC) - for frame checksum.
  • UU field (8-bits) -unused and can contain any value.
  • CPI field (8-bits) - unused, must be zero filled.

Datagram Encapsulation and IP-MTU Size:

• AAL5 uses a 16-bit length field, allowing up to 64K octets in a single packet.
• IP imposes an MTU of 9180 octets on ATM networks.
• If a datagram is larger than 9180 octets, IP fragments the datagram and passes each fragment to AAL5.
• AAL5 accepts, transfers, and delivers datagrams of 9180 octets or less.

Packet Types and Multiplexing:

• The AAL5 trailer does not have a type field, so each frame is not self-identifying.
• ATM has two approaches to sending and identifying information of a connection:
  • A virtual circuit is used specifically for a single high-level protocol (IP - no type needed).
  • Several high-level protocols can use the same virtual circuit (type field is required).

Packet Types and Multiplexing:
Single Protocol Transport:

• Advantage:
  • No additional information is required per packet to identify the high-level protocol.
• Disadvantages:
  • A separate virtual circuit is needed for each different high-level protocol (cannot share).
  • Most companies are charged per virtual circuit, so it is more costly.

Packet Types and Multiplexing:
Multiple Protocol Transport:

• Advantage:
  • Allows all traffic to travel over the same circuit, reducing costs.
• Disadvantages:
• Requires extra overhead (type field) in each packet to determine the high-level protocol.
• Packets from all protocols travel with the same delay and priority characteristics.

Packet Types and Multiplexing:

• The TCP/IP standard allows computers to access AAL5 using either transport method.
• If a Computer chooses to include the type field, it is required to use two specific link-layer headers:
  • IEEE 802.2 Logical Link Control (LLC) header, followed by the
  • SubNetwork Attachment Point (SNAP) header.

• LLC (24-bits) = AA.AA.03 (hexadecimal)
• SNAP (40-bits) - Two parts:
  • OUI (Organizationally Unique Identifier) - 3 octets. Identifies the organization that administers the values in the TYPE field. TYPE (2-octets) - identifies the packet type.
• For IP datagrams encapsulated over Ethernet:
  • OUI = 00.00.00 TYPE = 08.00

IP Address Binding in ATM:

• Encapsulating a datagram over ATM is straight forward, however address binding is difficult.
• ATM assigns addresses to all machines wishing to use a virtual circuit.
• ATM physical addresses are larger than IP addresses, so they cannot be encoded within them as static bindings.
• ATM does not support broadcasting (only multicasting via meta-signaling), therefore IP cannot use ARP for address resolution.
• Since PVCs are setup prior to use of the ATM network (by provisioning), each node does not know it's ATM physical address.
• They only know the circuit's VPI/VCI pair. This further complicates things.
• Switched virtual circuits (SVC) further complicate address binding because they require two levels of binding:

  1) IP address of the destination must be mapped to an ATM end-point address (hardware).

  2) Destination IP address must be mapped to the VPI/VCI pair for the SVC.

• Binding 2 is used each time a datagram is sent over ATM, binding 1 is done only once.

Logical IP Subnets:

• When a group of computers uses a large ATM network in place of a single local physical network, it forms a logical IP subnet (LIS) or Local area network ATM (LATM).
• This is used as a restricted form of address binding for larger neworks.
• Multiple logical IP subnets can be defined from all computers attached to the same ATM network.
  

• Each logical IP subnet functions like a separate LAN.
• All machines configured for a specific subnet exchange packets.
• Each LIS is treated by IP standard rules as a completely separate network.
• All LIS members share a single IP network ID, but it differs from all other LIS groups on the same physical network.
• All computers in a LIS can choose non-standard MTU sizes, but all must use the same size MTU on all LIS virtual circuits.
• Hosts communicating between separate LIS groups must use a router that participates among multiple logical subnets.
• No direct connections are allowed outside of LIS groups.

Connection Management:

• Creating a V.C., sending a single datagram, and then closing the V.C. is too expensive.
• Hosts must maintain a record of open virtual circuits so that they can be reused.
• Hosts use conventional IP routing when sending a datagram.
• The datagram and next-hop address are passed to the network interface which then examines it’s table of open virtual circuits to see if it can send it.
• If not, the network interface must:
  • Locate the next-hop computer's address.
  • Establish a switched virtual circuit.
  • Send the datagram using AAL5.
  • Add the circuit to it’s open circuit table.
• Logical subnets are justified by hardware and software constraints. Storing large tables of circuits, however can get costly and inefficient.

Address Binding within LIS
(LATM):

• Classic ARP relies on an efficient broadcast service from the physical layer.
• ATM does not offer a hardware broadcast mechanism, only a proposed multicast delivery scheme.
• ATM's point-to-multipoint connections offer bi-directional communication only between the originator and all destinations, but not among those destinations.
• To simulate broadcast, many point-to-multipoint connections would be needed, which is inefficient and not always possible.
• TCP/IP over ATM approach:
  • Use a dedicated service on each LIS (LATM) that maintains a cache of mappings.
• Mappings - are established through registration at boot time and periodically updated.

ATMARP and InATMARP:

• Address Resolution Protocol for ATM (ATMARP):
  • The same protocol as ARP, but with extensions to support functionality in a unicast server environment.
  • Given a network protocol address, it determines the associated ATM address.
• Inverse ATMARP:
  • Similar to the original RARP protocol.
  • Given a virtual circuit identifier (VCI) between two end-points, it determines the end-point's network layer and ATM address.
• Each LIS (LATM) must have at least one ATMARP server.
• Both protocols use messages similar to ARP and RARP, to provide registration and address resolution.

• The messages used include:
  • (In)ATMARP_REQUEST - request a service.
  • (In)ATMARP_REPLY - reply to a service request.
  • ATMARP_NAK - avoids ambiguity when a resolution cannot be made.
• Backup ATMARP service is desirable for an LIS since only one machine is responsible for all address resolution.

All work was written, produced, and is copyrighted by Daniel Z. Tabor Jr.
Page created by Daniel Z. Tabor Jr. Copyright ©1995 Illusion Industries Inc.