|
Although ATM can be seen as an extension to LAN switching,
ATM differs from LAN switching in a number of ways that give
ATM networks superior capabilities.
The following sections describe the functions and provide a
comparison of ATM and existing LAN technologies.
A Connection-Oriented Protocol
Traditional shared-media LANs use a connectionless protocol that
has proven adequate for most data-oriented applications.
Even though switched Ethernet and Token Ring are dedicated-media,
switched solutions,
they are still fundamentally connectionless in their operation
because they are based upon the original, shared-media protocols.
Currently, only ATM provides the connection-oriented environment
required for the emerging multimedia applications.
In addition the ATM environment offers considerable benefits for
running legacy applications that are connectionless.
Refer to Figure 2 for an example.
A network is a traffic-control system that manages the delivery of
goods to and from devices attached to the network.
Like the traffic-control system of a city that defines the rules
for the delivery of goods across its infrastructure of
streets and highways, each network protocol has its own set of rules.
Connectionless Protocols
Let us assume for a moment
that the traffic system of your city is based on
Ethernet's (or some other connectionless protocol's)
set of rules.
Each driver starts for a destination as soon as
the street appears to be clear of traffic.
The driver has no knowledge of the route or of others who might
also want to use the route at the same time.
It is a hit-and-miss system with congested traffic, collisions,
and restarts, with no guarantee of arrival times.
Connection-Oriented Protocols
Let us now assume that our
traffic system is based upon ATM's set of rules.
Drivers call ahead and request a route before the journey begins.
They get reserved routes for the trip, along with lanes
on the highway wide enough for their vehicles.
Each lane and its
width are reserved exclusively for the duration of the trip.
Nothing has been changed in the size of the highway (cabling
infrastructure).
The ATM traffic system automatically picks the most expeditious
route through the maze of streets and highways and continuously
reconfigures the routes as traffic patterns change.
In a connection-oriented environment, data is kept in the end-station
storage media until the connection to the receiving station
is made. Therefore the network is not burdened with the management
of data that is en route, thus allowing efficient operation
that is simpler, with predictable destination
arrival times. This is why ATM
has caused so much excitement in the industry.
Because all legacy LAN applications have been written for a
connectionless environment, it is necessary to map connectionless to
connection-oriented sessions to use the applications over ATM.
Two mapping techniques -- Classical IP and
LAN Emulation -- have been defined by industry standards
to facilitate interoperability among various vendors' products.
The migration path afforded by LAN Emulation might
actually slow down the
development of native ATM applications because the benefits that ATM
brings to the legacy applications are substantial.
Simply by moving the applications into ATM's connection-oriented
environment, legacy LAN applications run better because they
can take advantage of ATM's higher and dedicated bandwidth.
In addition to improved performance, inherent ATM characteristics such as
the ability to employ virtual LANs reduce the cost of operating and
managing legacy applications when they are run over ATM networks
using LAN Emulation.
Figure 2: Connectionless and Connection-Oriented Networks
Speed
In LAN switching, each frame has a different length and destination.
The processor in the switch must make an individual decision for every frame.
Therefore, the actual throughput capacity of a switch is directly linked to
its processor's power and limitations.
Techniques like cut-through switching, where
transmission is started as soon as enough bytes have been read to
recognize the destination address, can improve the latency
of the switch on a port-to-port basis.
However, filtering, port-speed adaptation
(10 Mbps to 100 Mbps, for example), and high error rates on the media
often prevent cut-through switching.
In ATM, the data is split into fixed-length cells of 53 bytes each,
where a header of 5 bytes contains the routing information.
The characteristics of the connection are negotiated ahead of time
and, if the network can guarantee the quality of service, the
call is accepted and the path is established.
Then, the cells are transmitted at hardware speed without the need to
reexamine the contents of the cell or perform intermediate
store-and-forward actions between the source and the destination.
A Multiplexing System
In a LAN environment (either shared or switched), applications in a
workstation or on a server take turns sending data onto the media.
A low-priority file transfer can delay the transmission
of a short frame that requires limited delay.
This delay will be repeated at every network node and
affect the performance of the network.
In ATM, because elements of information are split into 53-byte cells,
cells from different sources can be interspersed and queued according
to their individual priority.
Thus, fixed delays can be respected,
and quality of service can be set according to the application's
requirements rather than those of the adapter.
Superior Bandwidth Capability
ATM is, by its architecture, a full-duplex, switched solution. Although some
Ethernet and Token-Ring switches and adapters do have
full-duplex capability, the LAN switch must accommodate
diverse attachment port characteristics and will
act as a store-and-forward gateway between ports.
This reduces the real capacity and bandwidth of the network.
In ATM, bandwidth is a parameter in the definition of a
switched virtual circuit and is independent of the physical
attachment: there is no need for intermediate buffering.
In fact, if a physical link reaches capacity, additional connections can
be added to expand the bandwidth and support additional traffic.
Because this capability is one of the
fundamental building blocks for high-quality videoconferencing,
ATM networks not only provide better throughput for legacy applications,
they also provide the infrastructure for emerging applications.
Backbone Access
Most Ethernet and Token-Ring Switches are essentially multiport bridges.
They cannot use multiple uplinks, and the aggregate switch capacity must remain
commensurate with that of the uplinks.
Because ATM is a connection-oriented protocol, bottlenecks between the
workgroup switches and the backbone are easily removed by installing
additional uplinks between the workgroup switches and the higher
speed backbone.
ATM switches are able to set virtual circuits over diverse routes
according to the current network capacity usage or according to the
availability of a specific path.
This not only increases the possible link bandwidth but
also offers the possibility of bypassing a failing element.
As the number of users per floor increases in an
end-to-end ATM network, the bandwidth per user need not
be affected because of any limitation on uplink bandwidth.
Further, installing additional uplinks is simple and should cause little or no disruption
in the network. The ability to provide multiple links guarantees uninterrupted service to end users.
Quality of Service
ATM, with its multiplexing architecture, is designed to support traffic
with various bandwidth, jitter, and delay requirements.
This design feature allows ATM networks to support voice, video, and
data multiplexed on the same links.
Quality of service is established at the time that the connection
is made.
Implementing quality of service is dependent upon ATM being a
connection-oriented protocol.
The ATM Forum has defined four quality-of-service types that
are architected to handle the different types of traffic.
Constant Bit Rate (CBR) and Variable Bit Rate (VBR) are
particularly well-suited for supporting applications with stringent
requirements for quality of service, such as multimedia transmission
or high-quality videoconferencing.
Constant Bit Rate
CBR is a reserved bandwidth service.
A contract is established between the network and the end station.
The end station provides the network with parameters describing
the traffic for that specific connection at call setup time.
The network, in turn, allocates resources that match the parameters or,
if the resources are not available, rejects the call.
This is called call admission control.
Once the call is accepted, it is the end station's
responsibility to
send only traffic that is compliant with the contract.
The network checks the traffic against the contract, and noncompliant
cells are discarded.
Variable Bit Rate
Like CBR, VBR is a reserved bandwidth service.
The network allocates resources to the end station at call setup in
response to the traffic parameters requested by the end station.
However, in the case of VBR, in addition to a peak rate, a
sustainable rate and a maximum burst size are established.
The sustainable rate is the upper limit of the average rate, and the
maximum burst rate limits the duration of cell transmission at peak
rate.
These additional parameters allow the network to achieve statistical
multiplexing by allocating fewer resources for the connection than
would be required by the peak cell rate.
In most campus environments today, the majority of traffic is
data transfer that, for the foreseeable future, will operate over
ATM using either LAN Emulation or Classical IP mode.
These legacy applications are not able to specify the quality of
service that they will require.
The ATM Forum is proposing that this traffic employ either Unspecified
Bit Rate (UBR) or Available Bit Rate (ABR).
Unspecified Bit Rate
UBR is a non-reserved bandwidth service.
The cell loss ratio is unspecified, which means that the network is
not required to provide resources for a proposed UBR connection.
No flow control parameters are specified in the ATM Forum for UBR service.
Consequently, when UBR service is employed, cell discard seriously
impacts the overall performance of the system.
For example, a single cell discarded in a 192-cell packet (the default
size for an IP packet when using Classic IP over ATM) triggers
retransmission of the whole packet.
The network has transmitted 191 cells needlessly.
To avoid wasting network resources in this way, early packet discard and
partial packet discard can be implemented in any intermediate node
(switch) of the network. If a switch recognizes that a cell has been lost, it discards
the rest of the packet. If a sending station fails to acknowledge a congested condition,
the incoming switch in the network will reject packets until the congestion disappears.
When early packet discard and partial packet discard are implemented
in conjunction with virtual circuits, fairness and hop-by-hop
backpressure mechanisms ensure loss-free UBR operation.
Available Bit Rate
ABR service can be seen as a mix of reserved and non-reserved
bandwidth service.
Periodically, a connection polls the network and, based upon the feedback
it receives, adjusts its transmission rate.
Polling is done by Resource Management
cells sent by the source and looped back
at the destination so that the network elements and the destination
can provide feedback information.
In addition, network elements can create and insert RM cells in the
backward direction to provide feedback to the source more quickly.
Feedback can be explicit or implicit.
Explicit feedback specifies an explicit rate, while implicit feedback
indicates that congestion is either present or not present.
The source might receive explicit and implicit
feedback in the same RM cell.
ABR connections have a minimum guaranteed rate that cannot be
reduced by either explicit or implicit feedback.
The means by which the feedback is used to help
optimize bandwidth use is flow control.
Flow Control
In most implementations, including IBM's, CBR, VBR, and UBR traffic
is not subjected to flow control.
However, the forthcoming ABR quality-of-service type
defines an explicit flow-control mechanism based upon rate
control at the connection level.
The switches mark the circuit as a candidate for slowdown and notify
the applications causing the congestion to slow down.
This not only restricts the mean rate at which cells enter the
network, it also, when correctly tuned, removes the natural
burstiness of the cells' arrivals at a destination.
To the extent that cell inter-arrival time becomes more constant,
the mean waiting times at switches or other resources become
smaller.
The result is both increased fairness in network access and no lost
frames within the network, so response time and bandwidth
to the user are optimized.
In networks using LAN switches, the only flow control available is at
the link level and is proprietary.
It usually emulates flow control
by buffering stop and go, which adversely affects performance.
However, because flow control in an ATM network
is a characteristic of a
logical station in a virtual channel built end-to-end, it provides
superior control at the virtual circuit level.
Multicast Capability
In networks of LAN switches, filters can be used in the switches to
control broadcast traffic, but they have an adverse effect
on the overall performance of the network.
Multicasting capability and LAN Emulation, which builds broadcast trees
for Virtual LAN (VLAN),
are foundations of both video distribution and
videoconferencing and are exclusive features of the ATM architecture.
Unlike the recipients of
a broadcast message in a shared-media LAN, only those who
want the message will receive it.
Because traffic is connection-oriented, no network resources are wasted,
and there is no danger of a broadcast storm.
In VLAN implementations over ATM, multicasting and LAN
Emulation define precisely which stations should receive the
broadcast data.
In addition, the broadcast manager of LAN Emulation can be
augmented with filtering capabilities to reduce the amount of
overhead data generated by chatty LAN protocols such as AppleTalk.
Refer to Figure 3 for an example of an ATM
point-to-multipoint network.
Figure 3: ATM Point-to-Multipoint Network
Low Latency
The emerging bandwidth-intensive, isochronous applications can work nly
in an environment where the latency of any one switch is
predictable, constant, and extremely low as opposed to variable
and unpredictable.
In environments where variable-length data and per-frame
filtering are employed, latency is adversely affected.
In networks where ATM is employed end-to-end,
the transit time between any two points on the network will always be
the same, so the response time in a large network will be
predictable and constant.
Reducing Network Complexity
It is generally estimated that
up to 70% of the cost of network ownership is in
the cost of operating the network.
Therefore, the simpler the network is, the less
costly it is likely to be to operate.
If we look at a typical LAN environment today, shared-media LANs
are joined to backbones, often running a different LAN protocol, by bridges or routers.
Connection to the WAN is generally through routers as well. Bridges and
routers are high-maintenance items, especially in networks with many
moves, additions, and changes. Configurations have to be updated,
and the network has to be tuned for best performance.
ATM's quality of service and scalable bandwidth virtually eliminate the
need for network tuning.
Bridges and routers are replaced by simple connections between switches.
The result is a network that is more reliable, ready for
emerging multimedia applications, and that operates at a lower cost.
Refer to Figure 5 and Figure 4
for examples of a classical LAN structure and an ATM structure.
Figure 4. Classical LAN Structure
Figure 5. ATM Structure
Virtual LANs
In current LAN environments, workstations are tied to a port on a specific
device so that the functions available to that device correspond to
what the network administrator has predefined in the physical
network for security or for access to resources.
If the user relocates, the network administrator
must assign to the new physical port the
characteristics that match the user's need.
Because affinity groupings are often used as a way of managing
networks, when workgroups are reorganized, users have to be
reassigned to different physical ports.
In traditional LANs, if all or part of the affinity group moves
to a different building, the network administrator might have to make
physical modifications to the backbone devices (filtering tables or
interbuilding links) to preserve the previous capabilities.
VLANs, as implemented in ATM, allow users to belong to several
VLAN affinity groupings and share common services no
matter where they are physically located in the network.
Refer to Figure 6 to see an example of a VLAN structure.
In the VLAN environment,
who you work with becomes more important than where you work.
When end stations are using LAN Emulation, assignment to VLANs
is automatic and is provided by a LAN Emulation Configuration Server.
LAN Emulation guarantees assignments to the same VLANs,
regardless of the user's physical location.
VLANs, because they do not require the intervention of
the network administrator or the assistance of a technician to
enable and assign a LAN port, can be a major source of
cost savings in an environment with frequent moves.
Refer to Figure 7 for an example of LAN-to-ATM mapping.
Figure 6. VLAN Structure
Figure 7. LAN-to-ATM Mapping Concept
Network Access Control
Since the inception of LANs, designers and network administrators
have struggled to find methods of restricting access to only
authorized users.
Shared-media LANs use intelligent hubs to check MAC addresses against
the list of authorized users.
When a violation occurs, an alarm is sent and the port where the
violation occurred is shut down.
This protects the port but does not really
control access to the LAN.
Because Token-Ring and Ethernet switches have been designed for
performance, they do not perform address checking very efficiently.
Most implementations use some form of MAC address frame filtering,
which has to be performed on every frame in these connectionless
protocols.
Obviously, performance can be significantly degraded.
On the other hand, the inherent characteristics of ATM make protecting
the network from unauthorized users straightforward.
ATM's connection-oriented protocol requires a call to be
processed before any connection is established.
It is then a simple matter of implementation to check the connection
request against an authorization record.
This capability can be easily extended to legacy applications because
the LAN Emulation server will establish the connection for the
application.
If the registration is rejected, an alarm is sent and the station is
not permitted to use the network.
The port is not shut off, and the network performance is not degraded.
In addition, ATM allows you to implement other security measures, if
needed.
Comparison Table (Summary)
| Features | ATM | Switched Ethernet or Token Ring | ATM Customer Benefits |
| Bandwidth | 25.6 Mbps to 1.2 Gbps, full-duplex |
4, 10, 16, or 100 Mbps Single-attached stations can operate in full-duplex mode |
Can handle multiple streams of video and file transfer simultaneously |
|
| Backbone Access | Capable of multiple uplinks from local switch to backbone Capable of attaching multiple ports to a single server |
Single uplink Single link to a server |
As users and bandwidth requirements increase, ATM providesincremental capacity adjustments Higher availability |
|
| Quality of Service | ABR (per VC)
CBR (per VC)
VBR (per VC)
UBR (per VC) |
UBR only (at link only) |
High-quality video and audio sessions simultaneously with file ransfer |
| | Flow Control | End-to-end across the network, per session |
None -- inter-switch links are shared resources |
ABR with the flow control backpressure mechanism is the ultimate utilization of network resources |
| | Multicast | ATM Switch base Capability | None |
Provides the foundation for video distribution and conferencing |
| | Latency | Latency under 30 microseconds for any speed Latency-constant and filter-settable at call setup |
Variable latency of 40 microseconds to over 100+ microseconds for speed scaling 10Mbps to 100-Mbps store-and-forward requirement |
Critical functions or real-time applications like videoconferencing |
|
| Network Access Control | Connection-oriented protocol: station must establish the call before transmission Station must register to the switch Switch can check registration against a predefined list and deny access without impact to switch performance |
Any-to-any connectivity with no address checking Access filtering degrades performance Effective network control requires use of intelligent hubs on the microsegmented LANs | Lower operating cost Higher availability Access is independent of physical location |
|
| Virtual LAN | Allows for affinity groupings Logical implementation is independent of the physical boxes Virtual LAN assignments not tied to the physical port |
Virtual LAN assignment tied to a physical port on a specific witch; change requires modification at the switch level |
Flexibility in personnel movement Lower operating cost |
|
|