cagamosisthingyΔίκτυα και Επικοινωνίες

27 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

110 εμφανίσεις

Webster University



William C. Kratville


Distributed Systems

Submitted To: Douglas F. Gluntz

January 1999

Space Coast Campus

Palm Bay Florida



The use of ATM in a dis
tributed system enables it to support present and future

Needs for data, voice, and video communications.


Choosing ATM for networking in a distributed system allows for the use of


LAN and WAN technologi
es while making it ready for future technologies.

These future technologies include multimedia applications with real
time voice and video


Review of Literature

Topics covering ATM ranged from descriptions of the technology to its use in

current LAN and WAN environments. Most of the information was provided by

industry groups such as the ATM Forum and by manufactures of ATM equipment.


Information for this paper was obtained by searching the internet for ATM

re. Topics that included tutorials and white papers were of the most interest.


Analysis and Findings

Asynchronous Transfer Mode (ATM) is a good choice for networking in a

distributed system. It has the capabilities to emulate cur
rent LAN’s with LAN
Emulation (LANE) and to implement IP using techniques such as Multiple Protocols
Over ATM (MPOA).

Needs in a distributed system

Distributed systems have a need for multicasting capabilities and ATM
provides it with ATM Multicast
Servers (MCS) and ATM VC Meshes (, 1999).
Being a connection oriented technology true Quality of Service (QoS) is delivered on a
connection basis (Klessig, 1999). This QoS means that once a connection is
established it’s bandwidth and laten
cy characteristics are fixed.

ATM with it’s low latency makes it well suited for distributed systems. It
performs favorably with existing LAN technologies and should support interprocess
communications and client
server operations in these systems (Co
ulouris, 1996).

Origin of ATM

Synchronous transfer mode.

ATM evolved from earlier Synchronous Transfer Mode (STM) technology. STM
uses Time Division Multiplexing (TDM) to transmit packets of information. The


connections are made between switches an
d each channel has its own position or time
slot in the data stream being transmitted. If a connection is bursty this means that there
are empty slots being transmitted which represent lost capacity. It was decided these
empty slots could be used bursty
users and this is what ATM is designed to do
(, 1999).

Asynchronous transfer mode.

ATM is based on Asynchronous Time Division Multiplexing. The idea is to label
data packets according to there connection and not there place in the data stre
( 1999). ATM is asynchronous because the data packets can arrive at
irregular intervals but still arrive in order. There is not a set time when a machine will
transmit a packet. This means that packets enter the data stream whenever an e
mpty cell
is available. If no empty cells are available the packets are kept in a queue until a cell
becomes available.

Packet technology.

ATM is well designed for bursty traffic and allows operation between devices that
operate at different speeds
. It evolved from the packet switching and frame relay
technologies and integrates the multiplexing and switching functions
(, 1999).


ATM has been implemented in use with PCs, switched
Ethernet and Token
hubs, ATM
edge switches,

ATM enterprise network switches, and ATM
switches. It can be used as an end user service or as a networking infrastructure for other
services (, 1999).

ATM characteristics for distributed systems

Scalability and interoperab

With it’s scalability ATM allows for the implementation of multimedia
networking with real time voice and video. In most cases today, separate networks are
used to transmit voice, video, and data. ATM allows the use of one standard network for
ll three. ATM is a standards based technology providing interoperability between
networks. Many networks use one networking protocol on the local network and another
protocol on its long distance networks. With ATM the networks use the same consistent
standard (, 1999). ATM has been designed to be scalable in its
deployment across geographic distances, the number of users it will support, and in its
access and trunk bandwidths that range from the megabits to the gigabits.

ATM is not tied
to any specific physical transport. This means it is compatible
with currently installed physical networks. ATM can be transported over twisted pair,
coax, and fiber optic facilities (, 1999). It should be noted that
standard copper twis
ted pair will only support ATM in the LAN (, 1999).


Quality of Service.

The way ATM can guarantee quality of service is by means of Bandwidth on
Demand (BOND). Compared to other technologies including Ethernet, ATM does not
share the wire wit
h any other computers. In an Ethernet network all computers connected
to it wait for the wire to be silent before initiating a data transfer and monopolize it until
the transfer is completed. This effectively blocks others from being able to transmit unt
the previous user is done. In periods of heavy traffic this can cause a serious degradation
in performance for an individual user. Since ATM is connection oriented by establishing
a virtual circuit it will be able to give a constant performance to any

one user
(, 1999).

Connection oriented.

The result of ATM’s connection oriented behavior is that packets of information
are delivered in order. This contrasts with other forms of packet switching that will
deliver packets out of order re
quiring the destination to insure the correct re
ordering of


ATM and Ethernet

Ethernet is the most popular network technology being implemented for Local
Area Networks (LANs). Protocols such as TCP/IP that are running on Ethernet are
onnectionless. This means packets of information (datagrams) are passed from node to
node without any prior knowledge of the route they will take. The route between sender
and receiver can change as network conditions change. With ATM this is not the ca
ATM first establishes a route (connection) between the sender and receiver and then
sends all information over this established route until the transfer is completed.

The addressing for Ethernet and ATM are different in that the Ethernet requires a

48 bit (6 byte) address and ATM uses a 24 bit (out of a 5 byte header) address. Each
Ethernet Network Interface Card (NIC) has its unique 48 bit address built in and is
supposedly unique in the world. ATM does not use a unique address but relies on virt
channels (VCs) to establish connections. These VCs are held in tables of the nodes used
to establish the ATM connection (, 1999).

ATM Cells

53 bytes.

ATM information is made up of a fixed size packet of 53 bytes consisting of 48


of data and a 5 byte header . These fixed size packets are called cells. The fixed
cell size of ATM makes it possible for it to be used in time critical applications such as
voice and video (, 1999). The fact that 5 bytes out of every
bytes are used for header information translates to a 9.4% overhead penalty. This
overhead is sometimes referred to as “the cell tax”. The choice of 53 bytes was a result
of a compromise between networking organizations that wanted a 64 byte packet fo
r data
applications and telephony organizations which preferred 32 bytes of payload. If you split
the difference you get 48 (, 1999).

Cell types.

ATM cell types include the Idle Cell, Valid cell, Invalid cell, Assigned Cell, an
the Unassigned Cell. The Idle cell is inserted or extracted by the physical layer to
maintain the cell flow rate to the ATM layer . The valid cell has a header with no errors
or a header that was corrected by the Header Error Control (HEC) process. A
n invalid
cell has errors that were not corrected by the HEC process and are discarded at the
physical layer. An assigned cell contains user information and the unassigned cell is
transmitted when there is no information to be carried (, 1999
). The
unassigned cell should not be confused with the idle cell mentioned earlier.


Three layer architecture

ATM has defined three layers necessary for it’s implementation. They include
the ATM Adaptation layer (AAL), ATM layer, and the physical lay
er. The ATM and
Physical Layers provide no handshaking or acknowledgements of received data and there
is no mechanism for retransmission of lost or corrupted data. This allows the switching
and transport of cells to be very fast. Any reliability functio
ns must be provided by the
ATM adaptation layer (, 1999).

ATM Adaptation Layer.

The ATM Adaptation layer is divided into two sublayers called the segmentation
and reassembly sub layer (SAR) and the convergence sub layer (CS). The SAR is
used to
break down frames or packets of information into cells for transmission and then build
them back up when received. The CS provides the mechanism for mixing voice, video,
and data and provide the proper quality of service for it (Goralski, 1995).

The ATM adaptation layer (AAL) assures the appropriate service characteristics.
It places the data into the data bytes of the ATM cell. The size of the data bytes in the
cell depends on the AAL type and is usually between 45 and 48 bytes with 48 bytes b
typical. If the data to be transmitted is greater then the size of the cell the AAL will
divide the data and place it into subsequent cells for transmission. The AAL is the
interface in the ATM protocol stack to the higher layers of the protocol usin
g ATM.


There are four classes of service based on three parameters that are mapped into

six types of AAL.











Bit Rate








variable bit

oriented data



Type 1

Type 2

Type 3 and 4

Type 3 and 4

Type 5

Figure (Goralski, 1995).

Type 0 is a null type used for cell relay services that ar
e inherently cell based and

need no adaptation (Goralski, 1995). In this case there actually is no AAL since the data

is already in cell format.

AAL type 1 is used for isochronous, constant bit
rate services such as audio and
video. It contains 47 byte
s of payload per cell.

AAL type 2 is for isochronous variable bit
rate services such as compressed
video. It contains 45 bytes of payload per cell. There is an item type used to indicate
beginning of message, continuation of message, or end of message
. A length indicator
and cyclical redundancy check code is included.


AAL 3 and 4 are for variable bit
rate data such as LAN’s. They contain 44 bytes
of payload information and a multiplexing identification.

The AAL 5 contains 48 bytes of payload inform
ation and is referred to as the
Simple and Efficient Layer (, 1999).

ATM layer.

Most of the header information is generated in the ATM layer and contains fields
to deal with congestion, maintenance, and error control problems (bugs.wpi.ed
u 1999).

The ATM layer adds the 5 byte header consisting of the virtual path identifier, virtual
circuit identifier, and the payload type to the ATM cell and assures the cell is on the right

There are two types of ATM cell information us
ed in networking. One is for the
User Network Interface (UNI) and the other is for Network to Network Interfaces (NNI).
The difference between the cells is the NNI cell does not contain the Generic Flow
Control field but uses the extra data for a larger
virtual path identifier.

ATM Header.

The 5 byte header is designed for efficient switching in high
speed hardware.
Header information consists of payload
type information, virtual circuit identifiers, and


error checking information (
, 1999). The virtual circuit
identifiers consists of a Virtual Path identifier (VP) and a Virtual channel identifier (VC)
which together provide the information necessary for routing the cell across the network
(Coulouris, 1996).

The header is made up
of fields for Generic Flow Control (GFC), Virtual Channel
Identifier (VCI), Virtual Path Identifier (VPI), Payload Type (PT), Cell Loss Priority, and
the Header Error Control (HEC). The header is 5 bytes long and uses 4 bits for the GFC,
16 bits for the
VCI, 8 bits for the VPI when used as a UNI and 12 bits when used as an
NNI, 3 bits for the PT, and 1 bit for the CLP.

Cell loss priority.

The CLP allows for the discarding of packets in a congested network to insure
that other packets are processed a
ccording to their QoS parameters. The CLP is one bit
value specifying whether the packet is eligible for discarding or not (Goralski, 1995).

Generic Flow Control.

The GFC field is not needed in the NNI because no flow control is needed
between swi
tches. This allows the additional bits to be used for a larger number of VPIs.


Payload types.

The payload type field can be for idle cells, unassigned cells, physical layer
operation, administration, and maintenance (OAM) cells, and signaling cells.
payload type field is a 3 bit value allowing for 8 types to be defined.




User data, no congestion, SDU Type = 0 (AAL
5 Body Cell)


User data, no congestion, SDU Type = 0 (AAL
5 Body Cell)


User data, congestion, SDU Type = 1

5 End Cell)


User data, congestion, SDU Type = 0 (AAL
5 Body Cell)


Segment OAM F5 Flow Cell


End to End OAM F5 Flow Cell


Reserved for future traffic control and resource management


Reserved for future functions

Figure (Gora
lski, 1995).

Physical layer.

The Physical layer defines the electrical characteristics and network interfaces
used in ATM. The physical layer is not tied to any specific type of physical transport


(, 1999). If designed properly ATM ca
n run over any media
including T1, Ethernet, and FDDI.

The HEC is the only field not generated in the ATM layer. It is generated by the
physical layer and is used for error detection and correction of the header information.
The HEC can correct a singl
e bit error but can only detect if multiple errors exist. If a
packet is detected with multiple errors it is discarded (, 1999).

Service Classes

There are five classes of service currently defined by the ATM Forum UNI 4.0
specification (w, 1999). These classes include constant bit rate
(CBR), variable bit rate non real time (VBR
NRT), variable bit rate real
time (VBR
available bit rate (ABR), and unspecified bit rate (UBR).


Edge devices.

ATM is a co
oriented protocol as opposed to legacy LAN protocols that
are connectionless and require 48 bit addresses. Because of this difference Edge devices
are necessary to adapt existing network
layer protocols such as IP and IPX to ATM
(Klessig, 1999)
. The fact that ATM is connection oriented means a path must be
established from the source to the destination before data can be transferred.


In adapting existing networks to ATM the need for Edge devices arises. These
devices sit at the boundary
(or edge) of ATM networks and convert non
ATM traffic into
ATM cells (Klessig, 1999). The addition of a new traffic type requires only a new edge
device to convert it.

User Network Interface.

The ATM Forum maintains the standards used in ATM. The
ATM User Network
Interface (UNI) standard describes the size and structure of the packets used and how
connections are established (, 1999).

To connect from non
ATM devices on the edge of the ATM network the User
Network Interface (UNI) pro
tocol is used. This protocol dynamically signals for and
sets up Switched Virtual Circuits (SVC) across the ATM network between switches at
the edge. The ATM switches then use the Private Network Node Interface (PNNI)
protocol for setting up the path an
d routing (Klessig, 1999).

Network to Network Interface.

The Network to Network Interface (NNI) is used to connect separate ATM
networks together. The NNI describes how the networks communicate with each other.
The Private Network Node Interface
(PNNI) is the actual standard used to connect ATM


switches by different vendors. It is important that when selecting ATM hardware that
they have compatible versions of these standards so they may communicate with each
other (, 1999).

Protocol Over ATM

An important aspect of ATM is the Multi
Protocol Over ATM (MPOA)
specification. This allows ATM to be used for existing networks. The MPOA provides
for end to end network layer (layer 3) connectivity across an ATM network, the
on of heterogeneous IP subnets across both ATM and non ATM networks, the
direct connectivity between ATM devices below Layer 3, and insures interoperable,
distributed routing across the network. Some of the things that MPOA provides for are
LAN Emulation
(LANE), Next Hop Resolution Protocol (NHRP), and the Multicast
Address Resolution Server and Connection Server (MARSMCS) (www.cs.ubc.cs 1999).


Existing LAN’s can be converted to ATM using LAN Emulation (LANE). The
LANE provides a transition layer
between higher level connectionless protocols and the
lower level connection oriented protocols of ATM (Klessig, 1999). This transition layer
is on top of the AAL5 layer which is the simple and efficient layer of AAL. With LANE
no quality of service is
possible. Using LANE requires no change to the higher level
protocols that are currently being used with Media Access Control (MAC) protocols


using drivers such as NDIS or ODI . The implementation of LANE replicates most of the
existing characteristics o
f LAN applications. This allows these applications to run
transparently over ATM networks.

The LANE protocol handles connection and handshaking functions required by
ATM. It defines how the end stations communicate over ATM and how ATM attached
s communicate with clients on Ethernet and Token Ring LANs. Routable
protocols such as TCP/IP, Novell IPX, and DECnet, and non
routable protocols like
NetBIOS and Systems Network Architecture (SNA) can be handled by LANE
(, 1999). The LAN
E protocol allows the connection of ATM
networks with Ethernet and Token Ring LANs.

An ATM network can support multiple LANEs. Each one will operate totally
separate even if they are on the same physical network. Connecting between the LANEs
will sti
ll require bridges and routers. A LANE can connect separate LANs making in
look like one big LAN to its clients. There currently is no way to directly connect an
Ethernet emulated LAN to a Token Ring emulated LAN and still requires a gateway to
bridge be
tween them. The forwarding of packets between different emulated LANs
requires that a router be used. This router may be an ATM attached conventional router
or a form of ATM router implementing LANE at two or more interfaces to different
emulated networ
ks (, 1999).


Version 1.0 of the LANE specification guarantees that all standard
implementations of LEC will interoperate but servers from multiple vendors are not. The
next version of LANE should define server to server protocols allowing d
ifferent vendors
servers to work together (Klessig, 1999).

An emulated LAN consists of multiple LAN Emulation Clients (LECs) and the
LAN Emulation Service. The LAN Emulation service is made up of a LAN Emulation
Server (LES), a Broadcast and Unknown Serv
er (BUS), and a LAN Emulation
configuration server (LECS).

LANE communications.

All communications between LECs and between LECs and LESs is with ATM
virtual channel connections (VCC). There must be at least one control VCC between a
LEC and the LAN

Emulation service. More VCCs may be established for efficiency.
The VCCs can be either a Switched Virtual Circuit (SVC) or a Permanent Virtual Circuit
(PVC), or a mixture of both (Goralski, 1995).

There are two types of VCCs, control connections and da
ta connections. The
control connections carry administrative messages that include requests for initial
configuration and for addresses of other LECs. The data connections handle all other
communications and can consists of linking clients to each other
for data
direct unicast


communications, and linking clients to the BUS for broadcast and multicast
communications (Klessig, 1999).

The three types of control connections are Configuration Direct VCC, Control
Direct VCC, and Control Distribute VCC. The Co
nfiguration Direct VCC is a bi
directional point to point VCC set up by the LEC to the LECS. Control Direct VCC is a
directional VCC set up by the LEC to the LES. The Control Distribute VCC is a
unidirectional VCC set up from the LES back to the LEC
and is typically used in a point
to multipoint connection.

The three types of data connections are Data Direct VCC, Multicast Send VCC,
and Multicast Forward VCC. The Data Direct VCC is a bi
directional point to point
VCC set up between two LECs to excha
nge data. The Multicast Send VCC is a bi
directional point to point VCC set up between the LEC and the BUS. Finally the
Multicast Forward VCC is a unidirectional VCC set up to the LEC from the BUS and is
typically a point to multipoint connection (, 1999).

LAN Emulation client.

The LEC is responsible for the data forwarding, address resolution, and
control functions for a single end system in the Emulated LAN (ELAN). Hosts in a
emulated LAN are called LAN Emulation Clients. The LEC funct
ions as an end system.
It can be used on computers with ATM interfaces to operate as a file server. End user


workstations and personal computers can be a LEC. Ethernet or Token Ring switches
that support ATM networking can also be LECs as well as router
s and bridges
(, 1999).

Each LEC is identified by a unique ATM address. It is associated with one or
more MAC address reachable through the ATM address. A Network Interface Card
(NIC) would have a single MAC address where a LAN swi
tch would be associated with
all the MAC addresses reachable by the ports of the switch (, 1999).

The communications interface for the LEC is the LAN Emulation User Network
Interface (LUNI). It handles the sequence and contents of the message
s that the clients
use to transfer traffic on the LAN. The LUNI sits between the LEC or ATM host and the
network providing the LANE service. It provides the LES with initialization information
, registration information, and address resolution. The LUNI

moves data from the source
to destination via AAL
5 (Goralski, 1995).

LAN Emulation service.

The LAN Emulation service is made up of the LAN Emulation Server (LES), the
Broadcast and Unknown Server (BUS), and the LAN Emulation Configuration Service
CS). The LES is responsible for the control function for a particular Emulated LAN
(ELAN). Each ELAN has one logical LES that handles address resolution and control


information. The BUS provides multicast capabilities and the LECS assigns individual
E clients to the ELAN.

The LES and the BUS work together. They provide ATM with unicast and
multicast capabilities. The LES handles address resolution, control information, and
primary job is to register and resolve MAC addresses to ATM addresses. Th
e BUS is
designed to carry broadcast data and multicast traffic. The LES and BUS are typically
located on the same device (Klessig, 1999).

The LECS dynamically assigns different LECs to different ELANs. It provides
the clients with the address of the m
ost appropriate LES and maintains a database with
this information. A single LECS can manage the configuration information for a large
ATM network (Klessig, 1999). This means only one LECS is needed in an
administrative domain and can serve all ELANs wit
hin that domain.

Virtual LANs.

A virtual LAN is created when several ELANs exist on a switched internetwork.
The switched internetwork is made up of a combination of LAN switches on the ATM
backbone. The virtual LAN simplifies network management by al
lowing administrators
to make adds, moves, and changes by redefining groups in the network management
system (Klessig, 1999). A node on a virtual LAN could be made a member of a new


virtual LAN without changing its ELAN membership. This requires no phys
ical network

Native Mode Protocols.

A disadvantage of using LANE is the loss of ATMs quality of service (QoS)
capability. This is not the case with ATM native mode protocols. LANE deliberately
hides ATM so existing network protocols can be u
sed on the ATM network. In native
mode a minimal protocol stack is implemented to map network layer addresses directly
into ATM addresses and the network packets are then carried across the ATM network
(, 1999). The advantages of using LANE
is the higher layer protocols stay
the same for existing LAN applications. The use of ATM native mode will require a
change to these drivers to drivers that directly support ATM. With ATM native mode
QoS will be possible which is not possible with LANE.





LAN Emulation



Native Mode



Multicasting is when a host sends a packet or packets of information to a multiple
number of receivers. Multicasting using ATM is currently po
ssible using ATM
Multicast Servers (MCS) and ATM VC meshes. In order to support multicasting ATM
requires the use of multiple virtual circuits. This can lead to VC exhaustion when a large
number of multicasts take place.

In MCS the multicast packe
ts are first sent to the server and then sent to all the
receivers. The ATM VC mesh method has each sender set up a point to multipoint VC to
all the receivers (, 1999). The Multicast Address Resolution Server
(MARS) is used for address reso
lution and supports both MCS and VC meshes. The
MARS maintains a list of all local receivers in each group.


NHRP is the Next Hop Resolution Protocol. The NHRP consists of a NHRP
Server (NHS) that maintains a next
hop cache table with IP to ATM

mappings for all
nodes associated with the NHS. These nodes are attached to a Non
broadcast Multi
access (NBMA) network which permits multiple devices to be attached to it but does not
provide broadcast mechanisms (, 1999). All NHRP messa
ges are sent
using the IP protocol.


The NHS can be deployed in either server mode or fabric mode. In the server
mode each NHS within the NBMA is statically configured with the IP addresses of the
destination NHSs in the network. In fabric mode the NHSs

determine the destinations of
other NHSs using intradomain and interdomain routing protocols.

NHRS is used by a node when it needs to transmit across the NBMA network. To
resolve an address it sends an NHRP request to its NHS. If the destination is ser
ved by
this NHS the address is resolved and returned to the node requesting it. If the NHS is
unable to resolve the address it looks for another NHS in its routing table and forwards
the request to it. This process is repeated until the address is resolv
ed. When the NHS
resolves the address it then reverses the process traversing its way across NHSs until it
reaches the requesting node so it can establish a direct connection. This reverse
traversing allows all the NHSs in the path to learn the route and
cache the mapping so the
next time a node requests the same route the NHS can respond directly (,


References Cited (1999).

ATM Development. (1999).

ATM Packet Specification. (19
Asynchronous Transfer Mode (ATM) is Bad for Computer

Networks. (1999).
ATM Internetworking.

Coulouris, George (1996).
Distributed Systems Concepts and Design.

Second Edition.

Harlow, England: Addison
Wesley. (1999).

Unraveling Asynchronous Transfer Mode, Part Three.


Goralski, Walter J. (1995).
Introduction to ATM Networking.

New York: McGraw
Hill, Inc. (1999).

Asynchronous Transfer Mode.

Klessig, Bob (1999).

ATM LAN Emulation.
. (1999).

White Paper: Multi
Protocol Over ATM. (1999).

Connecting with ATM. (1999).

LAN Emulation (LANE). (1999).
Asynchronous Transfer Mode (ATM) Techn

ATM Primer.

Asynchronous Transfer Mode., (1999).

Nortel Networks: ATM Tutorial.




ATM adaptation layer


Asynchronous Transfer Mode


Broadcast and unknown server


LAN Emulat


LANE client


LANE configuration server


LANE server


LANE User Network Interface


Media access control


Permanent virtual circuit


Quality of service


Switched virtual circuits


User to network interface


virtual channel connections


virtual path connections