|
|
|
Managing Multivendor Networks
  
- 12 -
High-Speed Networking
Asynchronous Transfer Mode (ATM)
Networking
synchronous transfer mode (ATM) technology lends itself to applications with high
bandwidth requirements, such as video and multimedia. ATM not only enables the network
to ship huge amounts of data, it can also reduce use of the server. With an ATM configuration,
a NetWare server, for example, no longer has to wait for an Ethernet transmission
that would otherwise cause data to get backed up in the cache.
ATM networks are built on a star topology, with a centrally located ATM switch
and each desktop wired directly to the switch. ATM is a high-bandwidth packet-switching
and multiplexing mechanism. Network capacity is divided into cells of a fixed size,
which include header and information fields. These cells are allocated on demand.
This high-speed protocol will ultimately bring many advantages to wide-area networking.
However, the technology can be costly and might require other parts of the network
to be upgraded to handle the load. A server optimized for a 10Base-T network will
probably require upgrading to handle the increased amount of data flowing in from
the clients. Besides the servers, the clients might also need a hardware upgrade.
More than ever, computer networks are being pushed to their limits. Huge applications,
increased end-user demand for data, and high-demand applications such as videoconferencing
and multimedia are creating a need for more bandwidth than is often available on
a traditional 10 Mbps LAN. ATM, unlike Ethernet and token ring, is a connection-oriented
technology. In an Ethernet LAN, the amount of bandwidth available to each user decreases
as more people use the network. However, in an ATM network, the amount of bandwidth
available to each connection remains constant.
Earlier implementations of ATM used fiber optic cable and optical transceivers,
although commercial acceptance of ATM depends on its effective deployment on a variety
of media. ATM technology is rapidly moving towards the desktop level, and is now
available over Category 5 unshielded twisted pair (UTP) and Type 1 shielded twisted
pair (STP) cabling. UTP and STP are the most commonly used types of media in the
typical LAN environment. Category 5 UTP and Type 1 STP both support ATM transmissions
up to the full 155 Mbps. Cable lengths can reach 100 meters, and a maximum of two
connections per 100 meters is allowed.
Support for Category 5 UTP copper wiring means that ATM can now be brought to
the desktop in a manner that is transparent to end users. FORE System's (Warrendale,
Pennsylvania) PC ATM product line recognizes the need to bring ATM to the desktop,
and includes driver support for NetWare, Windows NT, and the Macintosh OS. In addition,
LAN Emulation techniques will permit existing applications running over NetWare,
Windows, DECnet, TCP/IP, MacTCP, and AppleTalk to run unchanged over an ATM network.
LAN Emulation also provides the means to establish internetworking between the ATM
and Ethernet or token ring LAN.
A recent LAN emulation specification, suggested by the ATM Forum, enables ATM
to be deployed in a LAN environment without having to change the system software.
In addition, the price is gradually decreasing on all fronts as competition increases
and new vendors enter the market. However, before ATM is widely accepted, more telephone
service providers must establish their ATM services, and ATM interfaces must be built
into network operating systems.
If ATM is brought to every desktop, every client gains the ability to send data
at speeds of 25 Mbps-155 Mbps, or more than 15 times the existing data rate of a
standard Ethernet LAN. The ATM architecture itself, however, has no upper speed limit.
ATM technology is still young, expensive, and lacking in standards, and an end-to-end
ATM network is still not a realistic possibility. It is used primarily to support
more specific, highly demanding applications that a traditional network would not
be able to support. A network with only ordinary, run-of-the-mill needs and productivity
applications running, for example, some database programs, productivity apps such
as word processing, spreadsheets, and e-mail, can run on a standard 10Base-T network
for quite some time without slowing down. Implementing an ATM network for these ordinary
tasks is like driving to the corner supermarket in an Indy 500 racecar.
ATM takes all types of traffic, including data, voice, and video, and transforms
it into 53-byte packets, which can then travel directly over a network via switching.
This small packet size lends itself to real-time applications, such as video. In
order to increase speed, the switches can route traffic through multiple paths. The
link, however, will appear as a point-to-point connection, or virtual circuit. Bandwidth
is available on demand, and users do not need to bear the expense of a dedicated
line.
Because of the lack of standards, various ATM switches are often incompatible.
The ATM Forum has done a considerable amount of groundwork for defining ATM standards,
however, and more vendors are starting to comply and offer complete ATM product lines.
An ATM solution can be costly by the time the switching equipment is paid for, workstations
are upgraded, and training has been planned. (In the near future, however, it is
likely that ATM will come to be accepted as a robust and complete backbone technology.)
The ATM Forum is a consortium of over 500 organizations. One of the first companies
to release ATM products was Fore Systems, one of the ATM Forum's principal members.
Fore released the first ATM adapter cards in 1991, the first ATM LAN switches in
1992, and remains the leader in this market. Fore approaches ATM with a four-tiered
architecture, as follows:
- Layer 1: ATM Transport Services. These services convert non-ATM traffic
to ATM cells, allowing all types of traffic to make use of ATM features.
- Layer 2: VLAN (Virtual LAN) Services. A VLAN is a logical association
of users with a common broadcast domain. VLAN technology permits a network to be
designed based on logical relationships, instead of physical connections.
- Layer 3: Distributed Routing Services. Although VLANs eliminate a substantial
amount of routing, some routing might still be required, such as establishing communications
between different VLANs, or conversion between different MAC types (Ethernet to token
ring).
- Layer 4: Application Services. This layer makes the services in the above
three layers available to existing applications.
An ATM network can carry three types of traffic: constant bit rate (CBR),
variable bit rate (VBR), and available bit rate (ABR). CBR accommodates
voice and video, and requires the ATM network to act like a dedicated circuit and
provide sustained bandwidth. VBR traffic is similar, except that the bandwidth requirement
is not constant. ABR traffic does not require a specified amount of bandwidth or
delay parameters, and is useful for most common applications such as e-mail or file
transfer.
The ATM network uses three techniques to manage traffic. They are as follows:
- Traffic shaping. This is performed at the user-network interface level
and ensures that the traffic matches the negotiated connection between the user and
the network.
- Traffic policing. This is performed by the ATM network and ensures that
traffic on each connection is within the parameters negotiated at the establishment
of the connection. An ATM switch uses a buffering technique called a "leaky
bucket" in order to police traffic. In the leaky bucket system, traffic flows
(leaks) out of a buffer ( bucket) at a constant rate, regardless of how fast the
traffic flows into the buffer.
- Congestion control. This is still being defined by the ATM Forum.
More on Congestion Control
The ATM Forum is still defining the congestion control technique of traffic management,
although two schemes have been proposed to control traffic flow, based on either
an end-to-end, or link-by-link basis. End-to-end schemes control the transmission
rate where the LAN meets the ATM device. The drawbacks of this method are that some
cells can be lost and it requires a considerable amount of buffer space. A link-by-link
flow control mechanism can support more users and uses less buffer space. This too,
has its drawbacks: it is more expensive and equipment to implement link-by-link control
is not commercially available. An integrated proposal, being considered by the ATM
Forum, would establish a default end-to-end mechanism, with an optional link-by-link
scheme.
For ATM to be widely accepted, however, switching systems must be capable of interoperating.
The ATM Forum's Private Network-to-Network Interface (PNNI) is a dynamic routing
protocol that can be used to build a multivendor ATM switching network. PNNI permits
different vendors' switching hardware to interoperate and establish a switched virtual
circuit (SVC) routing system. Under this model, several switches can work together
and act like a single switch. PNNI distributes information about network topology
between switches, so that paths can be calculated. It also provides for alternate
routing in the event of a linkage failure.
ATM Management
ATM networks, like traditional networks, need tools for analyzing and managing
switches and connections. However, these types of tools are in short supply for ATM
networks. As more software vendors respond to the demand, the availability of ATM
analysis tools will be another contributing factor to the widespread acceptance of
ATM. (A consortium led by Fore Systems has created a solution to the lack of management
tools for ATM networks. Fore proposes to extend Remote Monitoring, or Rmon,
to ATM networks, providing fault and performance monitoring services on ATM networks.)
Slow ATM
You might not need ATM if you don't have demanding applications like videoconferencing,
but you might still want more speed. AT&T Corp. is offering a new option, referred
to as "slow ATM." The service runs at 1.5 Mbps, instead of standard ATM's
minimum of 45 Mbps. The ATM Forum is, however, working on a standard for 25 Mbps
ATM. Either service would add extra speed over a standard network configuration,
while being less costly than standard ATM service. Many more low-end users could
be expected to move from 56 Kbps frame relay to the 1.5 Mbps service, rather than
moving immediately to high-speed ATM. The low-speed ATM network technology lets you
move gradually to high-speed ATM, as the need arises; this is an ideal solution for
easing into ATM technology without having to make a big commitment.
ATM and Frame Relay Internetworking
The ATM Forum and the Frame Relay Forum have jointly established a new standard--the
Frame relay to ATM PVC (Permanent Virtual Circuit) Service Internetworking Implementation
Agreement--to let users mix frame relay and ATM traffic on the same high-speed
network. This will permit frame-relay sites to move to higher-bandwidth ATM without
having to make an absolute choice between the two technologies. As a result, protocol
conversion software is unnecessary. The ability to use a mixed model permits a company
to use ATM at high-volume sites, while retaining frame relay at lower-volume sites
such as branch offices, and enabling the two to communicate.
If you use frame relay, but want to upgrade to ATM as a central hub, a hybrid
frame-relay/ATM internetworking service might do the trick. Protocols adopted by
the Frame Relay Forum and ATM Forum facilitate the establishment of such a hybrid
network. Under the service, the carrier provides protocol translations that enable
the ATM switch to talk to the frame-relay switch. The system lets you bring ATM into
an existing frame-relay network, instead of having to decide on deploying one or
the other.
FUNI
Frame relay to ATM internetworking provides for transparent linking of frame relay
sites to ATM sites. One way to achieve this is through a new standard known as the
Frame User Network Interface (FUNI), a service that performs a protocol conversion
between frame relay and ATM. This service permits a network manager to use existing
frame relay equipment, while gradually scaling up to ATM without having to make changes
to the existing frame relay network. FUNI is actually a low-speed, frame-based ATM
solution. The FUNI standard is still under development, while frame relay is widely
available and fairly stable. The difference between FUNI and frame relay is that
FUNI allows signaling and flow control to be extended to equipment on the customer
premises, and it might be an attractive solution for sites with many different applications
needing low-speed connections into an ATM network.
SNA Access to ATM
IBM is also working to support ATM in LAN/WAN environments. Price is one major
barrier to wide area ATM, but another is the amount of work required to interface
ATM with legacy networks. IBM's solution for joining ATM with its SNA/APPN installed
base uses the High Performance Routing (HPR) feature to provide native access
to wide-area ATM networks for SNA/APPN. SNA is well suited for interfacing with ATM
because of its service features. However, SNA routing is less suited to high-speed
networking. HPR overcomes these limitations. IBM's proposal is that the native interface
to ATM take place through the HPR feature. Under this model, mainframe SNA and APPN
would connect directly to ATM using either LAN emulation or Frame Relay emulation.
The APPN/ATM Internetworking specification, submitted by IBM to the APPN
Implementers Workshop, defines a method for SNA users to migrate existing applications
to ATM. The AIW is a consortium of vendors that includes IBM, Cisco Systems, and
3Com. The specification maps IBM HPR class-of-service routing to ATM's Quality-of-Service
specification. The specification will permit APPN/HPR users to make use of APPN's
class of service across an ATM net, without having to change existing APPC applications.
The APPN class of service defines route security, transmission priority and bandwidth
between session partners. HPR is an APPN extension that provides the ability to bypass
failures and eliminate network congestion. The specification would permit users to
deploy SNA class-of-service routines over an ATM net, without having to change existing
applications. IBM's HPR/ATM proposal is part of its strategy of helping users migrate
to switched network environments.
ATM Inverse Multiplexing
The ATM Forum is working on another way to ease the migration to IBM environments.
Their Asynchronous Transfer Mode inverse multiplexing (AIM) technique provides
for a more cost-effective deployment of broadband ATM over a WAN, by allowing a manager
to stay with their less expensive T-1 links as opposed to moving to a more costly
T-3 connection. T-3 runs at 45 Mbps, whereas a T-1 link runs at 1.544 Mbps. AIM establishes
a high-speed connection using multiple, point-to-point T1 links that are managed
collectively. The AIM specification permits ATM devices to be linked with a single
T-1 link; as the network requirements grow, additional links can be added, until
volume justifies the use of a T-3 link. AIM sends parallel streams across multiple
T-1 lines and dynamically balances the cells over all available links.
Quantum Flow Control
A consortium of vendors known as the Flow Control Consortium are proposing an
alternative to ATM, making it even more confusing for potential ATM users. The group,
which includes Digital Equipment and ten other companies, says that their Quantum
Flow Control (QFC) specification complements the ATM Forum's Traffic Management
Working Group's work on the Available Bit Rate (ABR) specification. QFC is designed
to interoperate with the ATM and Forum's Explicit Rate specification for ABR services.
LAN Emulation
LAN Emulation (LANE) defines how existing applications can run unaltered
on the ATM internetwork, and how the ATM internetwork itself can communicate with
Ethernet, token ring, and FDDI LANs. LANE, a specification of the ATM Forum, is an
internetworking strategy that permits an ATM node to establish connections to the
Media Access Control (MAC) protocol section on the Data Link Layer. This capability
permits most major LAN protocols to run over an ATM network, without having to modify
the LAN applications. LANE does this through three distinct techniques: data encapsulation,
address resolution, and multicast group management.
Each end station in the ATM network possesses a LANE driver, which establishes
the IEEE 802 MAC Layer interface. The driver will translate the MAC-layer addresses
to ATM addresses through the LANE Server's Address Resolution Service. Furthermore,
the MAC layer interface is transparent to high-level protocols, such as IP and IPX.
It is through this mechanism that a point-to-point ATM switched virtual circuit (SVC)
connection is established and data can then be transmitted to other LANE end nodes.
Multiple LANs can be emulated on a single ATM network, allowing for the creation
of virtual LANs (VLANs). A LANE driver located on an access device, such as
a router or hub, functions as a proxy for multiple end stations connected to the
device.
LANE offers advantages over a traditional LAN bridge environment, which is not
scalable enough to support a large internetwork. In addition, the LANE model supports
dynamic configuration, making it unnecessary to define physical connections and allowing
a host to be physically relocated, while remaining with the same VLAN.
Because existing 802 frame types are used in the LANE environment, an ATM adapter
can appear to an end station as an Ethernet or Token Ring card. Consequently, any
protocol that runs on Ethernet or token ring can also run on the ATM network.
The ATM Forum's LAN Emulation Over ATM 1.0 specification describes how
an end station communicates with the ATM network. The specification consists of two
parts: the LAN Emulation Client (LEC) and LAN Emulation Services. The
latter includes the LAN Emulation Server (LES), Broadcast and Unknown Server (BUS),
and LAN Emulation Configuration Server (LECS). The ATM Forum has gone out
of its way to demonstrate the computer industry's affinity for bizarre acronyms,
by collectively referring to this mechanism as the LAN Emulation User-to-Network
Interface (LUNI).
Despite the strange name, LUNI (pronounced "looney") goes a long way
toward providing multivendor compatibility. Through the LUNI specification, vendors
can easily establish interoperability between their various end stations.
Each ATM LAN end station has a unique MAC-layer address, as do standard 802 LAN
end stations. When one ATM end station is transmitting data to another ATM end station,
the first station will look for the second station's MAC address. After the first
station has discovered the second station's ATM address, any existing LANE connection
between the two can be used. If there is no existing connection, the first station
will initiate a connection using ATM signaling.
If an end station on an ATM LAN wishes to connect with an end station on an Ethernet
LAN, a few more steps are involved. Suppose John sits in front of a workstation on
an ATM LAN and wants to send the results of the World Series to Dan, whose machine
is connected to an Ethernet LAN. This is where the LAN Emulation Services (LES) come
into play. John's machine will send an address request message to the LES which sends
the request to a router on the Ethernet LAN. The router acts as a proxy LEC for the
end stations on the Ethernet LAN, and stores all the addresses of all the Ethernet
stations, including that of Dan's machine. When the address request is sent to the
Ethernet router, it is then broadcast to all of the end stations on the Ethernet
LAN. Dan's machine will eventually receive the request and respond to the router,
which then uses its own ATM address to make the connection.
Typically, connectionless LANs use bridges or routers to add additional end stations
to the internetwork. ATM, on the other hand, is connection-oriented, and data sent
between devices on an ATM network is seen only by the destination station. An ATM
network can use two types of connections: a permanent virtual circuit (PVC)
or a switched virtual circuit (SVC). The PVC is manually configured, where
the SVC is dynamically created by the ATM switch.
Also, the ATM network uses a different address structure from the connectionless
LAN. LANE takes care of the PVC and SVC connections transparently, using an address
resolution procedure to bridge the different addressing schemes and enable the two
to be connected. Products such as Fore Systems' ForeThought 4.0 include ATM Forum
LANE 1.0 software, which establishes a seamless connection between the ATM and Ethernet
LAN.
ATM LAN emulation mitigates much of the complexity of the ATM network, but is
only an interim approach on the road to full-scale ATM. Through emulation technology,
a shared-media LAN, such as Ethernet and Token Ring, can co-exist with ATM. This
permits a company to retain their original investments, while implementing a gradual
migration to ATM.
Multiple Protocols Over ATM (MPOA)
MPOA, an extension of the LAN emulation concept, is used to map network layer
addresses--such as IP or IPX--to ATM. Under an MPOA scenario, routing protocols such
as IP can use the ATM Quality of Service (QoS) features, with the ultimate
goal of allowing a LAN to work over ATM without having to migrate the LAN to native
ATM. As with LAN emulation, MPOA creates an ATM SVC (switched virtual circuit) whenever
a data relationship is established, creating a virtual router of sorts. This permits
network managers to create virtual subnetworks that go beyond routed boundaries regardless
of physical locations. The MPOA architecture is compatible with all routing protocols
capable of carrying addresses used by ATM, and is compatible with ATM's P-NNI specification.
There are three components to the MPOA architecture:
- Edge devices. These intelligent switches forward packets between legacy
LAN segments and the ATM infrastructure.
- ATM-attached hosts. Adapter cards that implement MPOA and enable the ATM-attached
hosts to communicate with each other and with legacy LANs connected by an edge device.
- Route server. This is actually a virtual server, not a physical device.
It permits the network-layer subnetworks to be mapped to ATM.
Frame Relay
Frame relay switching is a type of packet switching that uses small packets.
It also requires less"´rror checking than other packet switching mechanisms;
instead, it relies more on end user devices, such as routers or front-end processors,
to provide error correction. Frame relay is similar to X.25 in that it is a bandwidth-on-demand
technology. It establishes a pool of bandwidth which is made available to multiple
data sessions sharing a common virtual circuit.
The Frame Relay Implementors Forum, a consortium that includes Cisco Systems,
Digital Equipment, Northern Telecom, and StrataCom, has established a common specification
for frame relay connections. The specification is based on the ANSI frame relay standard
and includes an extension that establishes a local management interface.
In the past, frame relay networking technology was used only in large WAN environments,
although it is coming to be used as a tool to carry multiple types of traffic, including
data, fax and even SNA traffic. It is less costly than a dedicated private line solution,
and extremely fast. Frame relay offers a number of benefits. SNA over frame relay
adds savings by enabling users to eliminate private lines typically used to support
critical applications. A high-speed frame relay network will let users transmit data
at a rate of 1.544 Mbps.
Voice Support and NNI
Although it does not currently support voice transmission, the potential of voice
support is tantalizing. Frame relay voice support would let you make voice calls
on the frame relay net, potentially saving big money on international calls.
However, Network-to-Network Interfaces (NNI) have not yet been sufficiently developed.
NNIs are used to let carriers interconnect their separate networks, and are an essential
part of international frame relay.
Frame relay technology is becoming much more attractive economically, and carriers
are getting intensely competitive. In many circumstances, frame relay is superior
to a private line for data networking scenarios. The carriers' pricing models should
be taken into account when considering a frame relay solution. Pricing schemes are
complex, and include port charges for physically connecting to the network, charges
per PVC (permanent virtual circuit), and charges for local access. Other charges
include COC (central-office connection) tariffs, which cover the cost of the
connection between local access service and the interexchange carrier.
The lack of switched virtual circuit (SVC) services has delayed the widespread
implementation of frame relay in the past. However, manufacturers and service providers
are starting to implement these services in earnest. The lack of SVC services caused
customers to instead rely on frame-relay PVCs. SVCs would permit a network manager
to establish a frame relay connection on demand, and replace the need for PVCs between
sites.
Software is starting to become available to integrate voice, fax, and data networks
over frame-relay. Products are available to enable a frame relay network to handle
all three types of traffic. This type of software would naturally give priority to
voice traffic, sending it at a Committed Information Rate--which reduces delays typically
associated with sending voice over frame relay.
Switched Multimegabit Data Service
(SMDS)
Switched Multimegabit Data Service (SMDS), a connectionless service, can
be advantageous in some multivendor networks over ATM or frame relay technologies.
Network design under an SMDS architecture is actually quite simple. With frame relay,
on the other hand, you have to assign and configure PVCs (permanent virtual circuits)
between locations. ATM has similar complex design requirements. SMDS, on the other
hand, establishes any-to-any connectivity. Each location has its own E.164 address,
so all you have to do is assign it a port connection speed. After a site is hooked
up, it can communicate with any other site on the SMDS net.
SMDS is a scalable solution, and is capable of keeping pace with an increased
number of sites at a low incremental cost. SMDS port speeds are also scalable. In
addition, the ATM Forum and SMDS Interest Group have established a specification
for internetworking SMDS and ATM services. SMDS networks have a group addressing
feature, which can be used to create multiple virtual private networks that can be
easily modified as needed. However, it is limited to data only, and is not suited
for real-time multimedia as is ATM.
Fibre Channel
The ANSI Fibre Channel standard offers higher available bandwidth than
ATM, and more products supporting Fibre Channel are available in the marketplace.
Sun and HP both have workstations that support Fibre Channel networks. ATM was designed
as a cell-based, high-speed network architecture for data and voice traffic. Fibre
Channel, on the other hand, is a high-speed architecture for connecting network devices,
such as PCs and workstations, and high-speed hardware (such as hard drives) that
are usually connected directly to a system bus. The bus (channel) offers the combination
of high transmission speed with low overhead. The standard supports four speeds:
133 Mbps, 266 Mbps, 530 Mbps, and 1.06 Gbps. Fibre Channel NICs supporting these
speeds are currently available. ANSI has approved 2.134 Gbps and 4.25 Gbps Fibre
Channel specifications (although the technology for these rates have not yet been
made commercially available). Commercially available ATM products, on the other hand,
usually support only the middle of the ATM transmission rate range.
Switching in Fibre Channel networks is done by ports logging directly onto each
other, or to connecting devices (the "fabric"). Fibre Channel architecture
consists of five layers:
- FC-0. This is the physical layer, and includes the Open Fibre Control
system. If a connection is broken, Open Fibre Control permits the receiving device
to change over to a lower-level laser pulse.
- FC-1. This is the transmission protocol layer,
- FC-2. This is the Signaling Protocol layer. FC-2 defines three service
classes: Class 1 is a dedicated connection, class 2 provides for shared bandwidth,
and class 3 is the same as 2 except that it does not confirm frame delivery.
- FC-3. This layer defines common services.
- FC-4. This layer includes the Upper Layer Protocols (network and channel protocols).
High-Performance Parallel Interface
(HIPPI)
Fibre channel is meant to be the successor to HIPPI (high performance parallel
interface), which was developed to connect heterogeneous supercomputers with
IBM mainframes. Like HIPPI, the primary application for fibre channel has been clustering,
or joining processors together in a point-to-point link for parallel processing.
It can also be used to link the processor to a storage array. The advantage of frame
relay over HIPPI is that processors can be located several kilometers apart, whereas
HIPPI had a much shorter maximum distance (at least during its earlier incarnation).
Fibre channel is not currently used as a LAN backbone technology (although it is
being proposed for that purpose).
Is Fast Ethernet still not fast enough? Although 100Base-T, ATM, and other fast
networking technologies are probably more than most people need. Some areas, such
as scientific visualization, fluid dynamics, structural analysis, and even cinematic
special effects, require a gigabit-per-second throughput. HIPPI, a connection-oriented,
circuit-switched transport mechanism, offers an incredible data rate of up to 1.6
Gbps. Originally designed in the late 1980s as a supercomputer technology, the latest
incarnation of this ANSI standard is now applied to workstation clusters and internetworks.
Although it is limited to a distance of 50 meters in a point-to-point connection
over copper wire, it can reach 300 meters over multimode fiber, and up to 10 kilometers
over single-mode fiber. In addition, the original specification has been extended
to allow the 50 meter copper wire connection to be extended to 200 meters by cascading
multiple switches.
Much has been done to extend the capabilities of HIPPI below the supercomputer
level; it can now be applied to an Ethernet internetwork or workstation cluster.
HIPPI works well with most LAN and WAN technologies, including all varieties of Ethernet,
FDDI, ATM, Fibre Channel, and standard TCP/IP protocols. It is capable of linking
workstations and other hosts, and connecting workstations to storage systems at very
high speeds. While HIPPI offers greater potential than other high-speed technologies
such as ATM, HIPPI can coexist well with an ATM network, combining ATM's wide-area
possibilities with the super high speed throughput of HIPPI over the local area.
(HIPPI-ATM interfaces are still under development by the ANSI committee and HIPPI
Networking Forum. Such a connection would encapsulate HIPPI data, send it over the
ATM network, and then rebuild it at the other end.)
Fast Ethernet
The Fast Ethernet specification provides ten times as much bandwidth as
a traditional 10Base-T network. Some consider the technology to be overkill, especially
for smaller networks running standard productivity applications. Very few corporate
users even use more than a few Mbps of bandwidth, and do well with their existing
Ethernets. However, there are cases in which 100Base-T and other fast networking
scenarios are practical and economical. Fast Ethernet networks might prove invaluable
to professionals in the fields of engineering, CAD, and multimedia. Using Fast Ethernet
as a backbone in a client/server network might make sense, especially if a high number
of clients want to access the backbone network.
100BaseT is an extension of the IEEE's official 802.3 Ethernet standard. The 100Base-T
network interface cards are fairly easy to install and widely available, and use
standard two-pair UTP wiring (category 3, 4 or 5). Chances are, you already have
category 3 or 4 wiring in the walls, which makes upgrading to 100BaseT fairly economical.
There are actually three physical layers to the 100Base-T specification:
- 100Base-TX. The most common layer, 100Base-TX is full-duplex capable but
supports only category 5. Most Fast Ethernet products target category 5 installations
only.
- 100Base-T4. This is a four-pair system for category 3, 4, or 5 UTP cabling.
100Base-T4 can be more difficult to install and maintain because it requires four
pairs of wiring, and there are fewer products available.
- 100Base-FX. This is a multi-mode, two-strand fiber system. Use of fiber
optic cable yields a maximum distance of 2 kilometers.
All three types of systems can be interconnected through a hub.
Hybrid 10/100 Mbps network interface cards (NICs) can run $100 more than straight
10 Mbps cards, (although prices are likely to come down when the market for Fast
Ethernet matures). These hybrid cards are usually software-configurable and capable
of running at either speed. They can also include an auto-negotiation feature, which
is a technique used by the card to communicate with the hub to automatically determine
the environment. It will automatically sense whether it is 10 Mbps, 100 Mbps, half-duplex,
or full-duplex. Some Fast Ethernet products might permit cables for both 10BaseT
and 100BaseT networks to be directed to a single hub.
Despite advancements in 100Base-T, 10Base-T is still the most widely used network
infrastructure, typically implemented in a star configuration with a central hub.
However, as demand for data increases and applications grow in size, high-speed LANs
are gradually becoming more important. Technology such as Fast Ethernet can provide
the faster response times that impatient end users need, as well as the additional
bandwidth that is required by high-end applications.
The Fast Ethernet standard has become the predominant standard for high-performance
networking. Like 10Base-T, 100Base-T is based on the Media Access Control (MAC) protocol
section of the Data Link (Layer 2) section of the OSI model. As a result, 100Base-T
can be easily integrated into an existing 10Base-T network and run over existing
cabling. Because many vendors now support 100Base-T with new products, including
hubs, routers, bridges and interface cards, Fast Ethernet networks enjoy a high level
of multivendor support. Adding 100Base-T to an existing 10Base-T network can be a
gradual process and is often largely determined by existing cabling. As new stations
are added to the network, dual-speed 10/100 adapters can be installed in anticipation
of full migration.
Data can move between 10Base-T and 100Base-T stations without protocol translation
because Fast Ethernet retains the same protocol as plain Ethernet--Carrier Sense
Multiple Access Collision Detection (CSMA/CD). A simple bridge will carry out
this movement between 10Base-T and 100base-T. Migration from 10Base-T to 100Base-T
is quite simple, because of the high level of compatibility and because it is based
on the same technology and protocols. Most 100Base-T NICs are actually 10/100 cards
and can run at either 10 or 100 Mbps. Many cards are auto-sensing and will automatically
detect whether it is connected to a 10Base-T or 100Base-T hub.
An alternative to 100Base-T is 100VG-AnyLAN. This 100VG technology eliminates
packet collisions and provides for more efficient use of network bandwidth. The 100VG
also provides some facilities for prioritizing time-sensitive traffic. Despite these
technical advantages, many network professionals still prefer 100Base-T simply because
it is more familiar--it uses many of the same access mechanisms found on standard
10Base-T nets. However, being based on the same mechanisms means that 100Base-T is
not suitable for time-sensitive or real-time applications, such as videoconferencing.
Gigabit Ethernet
Gigabit Ethernet is the next step in the evolution of Ethernet. This wondrously fast
gigabit-per-second Ethernet technology is still a long way off, and is currently
little more than vaporous talk coming out of standards committees. However, this
promising technology is likely to be less expensive than ATM and more scalable, not
to mention less expensive to deploy because the costs normally associated with frame
conversion are absent. The IEEE 802.3 working group studying Gigabit Ethernet might,
if all goes well, have a specification by 1998. Under the group's initial design,
Gigabit Ethernet would retain 100Base-T's frame size and CSMA/CD scheme, but would
use the physical layer of the Fibre Channel architecture as underlying transport
mechanism.
© Copyright, Macmillan Computer Publishing. All
rights reserved.
|
