Comparison of Two Shortcut Routing Schemes: 
Multi-Protocol Over ATM vs. Fast IP


1 Table of Contents

1	Table of Contents			1
2	Executive Summary			1
3	Background				2
4	Architecture Details			2
4.1	Multi-protocol Over ATM (MPOA)		2
4.1.1	LAN Emulation (LANE)			3
4.1.2	Next Hop Resolution Protocol (NHRP)	4
4.1.3	How MPOA uses LANE and NHRP		4
4.2	Fast IP					5
5	Comparative Analysis			5
5.1	Quality of Service			6
5.2	Price / Performance Benefit		6
5.3	Migration Path / Standards		6
5.4	Protocol Support			7
5.5	Scalability				7
5.6	Filtering / Firewalls			8
6	Conclusions				8
7	References				8

2 Executive Summary

During the past year a number of schemes have emerged to solve the
router bottleneck problem by avoiding 
or minimizing packet by packet handling of data by routers.  In this
paper we study two such schemes: 
ATM Forum's Multi-Protocol Over ATM and 3Com's Fast IP.  Both MPOA and
Fast IP are flow based 
schemes, i.e. shortcut is set up based on the volume of traffic going to
a particular end station.  Both MPOA 
and Fast IP are primarily meant for campus LAN environment.  MPOA is an
ATM only solution while Fast 
IP also works with packet switched networks.  MPOA is based on two core
technologies, LANE and 
NHRP. LANE provides emulated LANs that emulate the services of Ethernet
and token ring LANs across 
an ATM network.  NHRP is an address resolution protocol that permits
queries between different emulated 
LANs. NHRP allows MPOA devices in one subnet (VLAN) to set up direct
(layer-2) connections to 
MPOA devices in another subnet thus avoiding hop by hop handling of
packets by routers.  Like MPOA, 
Fast IP also uses a variation of NHRP to enable two systems on different
VLANs to communicate directly.

Both MPOA and Fast IP currently do not have any explicit support for
Quality of Service, however both 
have plans in place for future support.  There are some concerns about
the scalability of these schemes to 
WANs which haven't been fully addressed yet. Both these schemes offer
significant cost benefits in the 
campus LAN environment as they can simplify inter-VLAN communications
and reduce the role of 
routers, which should translate into a need for fewer router ports.  In
both these schemes no changes are 
required at the core of the network and only the edge devices are
affected.  Initially both these schemes will 
only support IP, however they can be extended to use any other network
protocol.  Networks with filtering 
and firewalls have to do the "policing" during the setup of the shortcut
as there is no way to monitor the 
network once the shortcut has been established.

MPOA , Fast IP, and other shortcut routing schemes are in their
evolutionary stages and it will be some 
time before these schemes stabilize and industry has enough experience
to determine which scheme lives 
up to its promise.

3 Background

Over the last year a number of schemes have emerged to tackle the issue
of how to deliver the functions 
traditionally associated with routers in a way that's optimized for
switched networks.  The goal is to exploit 
the speed of switches and control function of routers.

In networks where layer-2 switches operate at a very high speed, layer-3
slow routers can be a potential 
bottleneck.  To solve this router bottleneck problem many vendors
initially introduced schemes that blurred 
the lines between layer-2 switches and layer-3 routers by providing LAN
switches with on-board routing.  
However during the past year a number of schemes have emerged that take
a different approach by 
avoiding or minimizing packet by packet handling of data by routers.  

In spring of 1996 Ipsilon Networks introduced IP switching which
exploits ATM switching fabric with a 
new set of IP-oriented protocols. This scheme has a potential of
delivering millions of packets per second 
as compared to hundred of thousands per second throughput of traditional
routers.  In response to Ipsilon's 
IP switching several vendors introduced their own solutions all based on
the idea of minimizing handling of 
data packets by routers.   Although the ATM Forum has been working on
its own solution for some time 
the announcement by Ipsilon forced them to speed up their work and they
came up with the scheme known 
as Multi-Protocol over ATM (MPOA). Similarly Cisco Systems response to
IP switching is Tag switching 
whereas 3Com has its own solution known as Fast IP.

In this paper we will look at MPOA in detail and compare it to the
3Com's Fast IP.  We are comparing 
these two schemes because of their similarities in that both MPOA and
Fast IP can simplify inter-VLAN 
communications at the campus LAN level and reduce the role of routers.
Fast IP uses a core protocol 
(NHRP) used in MPOA.

4 Architecture Details
4.1 Multi-protocol Over ATM (MPOA) 

Multi-protocol Over ATM (MPOA) is an address-resolution scheme whereby a
sending station discovers 
the ATM address of either the end host or an edge device closest to the
destination host so that it can 
establish a switched connection to the destination bypassing the routers
on the way.

MPOA supports both layer-2 and layer-3 connectivity between two
ATM-attached devices across an ATM 
fabric. Hence it enables both bridging and routing information to be
used to connect to the device closest to 
the destination.  For bridging, MPOA uses a Revised Version of ATM
Forum's LAN emulation 
specification (LANE).  LANE allows the bridging of traffic between
ATM-attached devices on the same 
subnetwork as well as supports emulated LANs (VLANs). For routing, MPOA
uses the Next Hop 
Resolution Protocol (NHRP) which is a request response mechanism that
allows a source station to 
determine the network layer (Layer-3) address of the destination device.

The primary goal of MPOA is the efficient transfer of inter-subnet
unicast data in a LAN environment. 
MPOA integrates LANE and NHRP to preserve the benefits of LAN Emulation,
while allowing 
intersubnet, network layer protocol communication over ATM virtual
channels without requiring routers in 
the data path. MPOA provides a framework for effectively integrating
bridging and routing with ATM in 
an environment of diverse protocols, network technologies, and VLANs.
This framework is intended to 
provide a unified model for overlaying network layer protocols on ATM. 

MPOA allows the physical separation of network layer route calculation
and forwarding, a technique 
known as Virtual Routing.  Virtual routers are a set of MPOA devices
operating over an ATM fabric that 
collectively provide the functionality of a multi-protocol router.  MPOA
splits the routing functionality 
between the route server and edge devices.   Edge devices examine the
destination address of packets 
received on legacy LAN segments and decide how to forward those packets.
If the packet doesn't need to 
go outside the subnet, the work of the edge device is done ? it merely
bridges the packet, using LAN 
emulation to resolve the ATM address and establish a virtual circuit to
the destination.  If the packet must 
be routed, the edge device examines the destination network layer
address of the edge device and looks up 
the ATM address corresponding to that network layer address.  The edge
device then establishes a direct 
virtual circuit to the appropriate destination.  The edge device gets
the destination ATM address from the 
route server, which can use various protocols to discover the ATM
address of any device in the network.  
To minimize the number of times an edge device visits the route server,
edge devices also maintain their 
own cache of addresses. This separation of edge devices, which forward
packets, and route servers, which 
supply routing information, has the following key benefits:
? It allows efficient inter-subnet communication.
? It increases manageability by decreasing the number of devices that
must be configured to perform 
network layer route calculation.
? It increases scalability by reducing the number of devices
participating in network layer route 
calculation.
? It reduces the complexity of edge devices by eliminating the need to
perform network layer route 
calculation.

The goal of MPOA as mentioned above is to remove the router bottleneck
problem.  It achieves this by 
allowing MPOA devices in one subnet (VLAN) to setup direct connections
with MPOA devices in another 
MPOA subnet thus obviating hop by hop handling of packets by routers.
The support for VLANs is 
inherent in the design of MPOA since the ATM fabric is treated like a
cloud between the source and 
destination devices which allows the layer-3 topology to differ from
layer-2 topology. This allows network 
managers to build virtual subnets so users can be grouped together as
part of a virtual network regardless of 
where they are physically located in the network.

The next two sections give a brief description of MPOA's core
technologies, LANE and NHRP.

4.1.1 LAN Emulation (LANE)

The ATM Forum's LAN Emulation (LANE) provides emulated LANs that emulate
the services of Ethernet 
and token ring LANs across an ATM network. LANE provides many benefits
including interoperation with 
Ethernet and token ring hardware and software, allowing a subnet to be
bridged across an ATM/LAN 
boundary. LANE allows a single ATM network to support multiple emulated
LANs. By utilizing emulated 
LANs, network layer protocols may operate over an ATM network in
essentially the same way that they 
operate over Ethernet and token ring LANs. LANE works at the MAC layer
and enables legacy Ethernet, 
token ring, or FDDI traffic to run over an ATM network with no changes
to applications, operating system, 
or network adapters.  

In version 1.0 of the specification, the ATM Forum defined how a LAN
emulation client (LEC) interacts 
with a LAN emulation service across the User-to-Network Interface (UNI).
LANE sits above the ATM 
adaptation layer (AAL). It masks the connection setup and handshaking
functions required by the ATM 
network from the higher protocol layers, enabling LANE to be independent
of upper-layer protocols, 
services, and applications. It also maps the MAC address based data
networking protocols, such as Ethernet 
and token ring, into ATM virtual connections so that the higher-layer
protocols think they are operating on 
a connectionless LAN.

An emulated LAN consists of multiple LECs communicating through the LAN
emulation UNI (LUNI) to a 
single LAN emulation service. A LEC is a combination software and
hardware agent within a networking 
device that handles data forwarding, address resolution, packet-to-cell
segmentation and reassembly (SAR), 
signaling, and other control functions. Each network component,
including servers, can support multiple 
instances of a LEC, allowing multiple emulated LANs to exist
simultaneously on the same physical 
network. 

The LANE service consists of a LAN emulation server (LES), a broadcast
and unknown server (BUS), and 
a LANE configuration server (LECS).  Unlike legacy LANs, which support
multipoint-to-mulitpoint 
broadcast, ATM supports only point-to-point (unicast) and
point-to-multipoint (multicast or broadcast) 
connections. The LES and BUS, which are typically co-located, work
together to transfer unicast and 
broadcast traffic. The LES handles address resolution and control
information. Its primary job is to register 
and resolve MAC addresses to ATM addresses. The LES also enables LECs to
join an emulated LAN and 
register/unregister multiple LAN destinations.

The BUS is designed for carrying broadcast data, such as TCP/IP address
resolution broadcasts. It also 
handles all multicast traffic and broadcasts the initial unicast frames
sent by the LEC, while the LES works 
in tandem to provide the appropriate ATM address for establishing a
direct virtual channel connection, 
either switched (SVCs) or permanent (PVCs). The LECS provides
connectivity for LECs so they can 
obtain configuration information, and assigns each LEC to a LES.

4.1.2 Next Hop Resolution Protocol (NHRP)

NHRP is an address resolution protocol that permits queries between
different subnetworks.  It reduces 
extra router hops required by traditional IP routers.  A source station,
such as a host or a router, connected 
to a MPOA subnetwork can use NHRP to determine the layer-3 address and
ATM address of the next hop 
towards a destination station.  Like LANE, NHRP is based on a
client-server model.  An NHRP client 
initiates NHRP requests to obtain NHRP services. NHRP server provides
NHRP services within the ATM 
cloud.

NHRP works as follows. Station "A" (which could be the source or an edge
device) wants to send data to 
station "B" and so needs to resolve station B's layer-3 (IP) address to
an ATM address.  If B's address 
resolution information is in A's cache it uses that, otherwise A
constructs an NHRP request packet and 
emits towards the NHRP server.  When the NHRP server receives the packet
it checks to see if it serves 
station B. If it does not server station B it forward the request to
another NHRP server.  The NHRP server 
which serves station B resolves B's ATM address and generates a positive
NHRP reply on B's behalf. If B 
is not on the ATM network, the next hop layer-3 address will be that of
the nearest edge device/router to B 
through which packets for station B are forwarded.

4.1.3 How MPOA uses LANE and NHRP

Devices acting as MPOA clients will initiate communication using LANE.
During the normal LANE 
address resolution process, a LANE client uses the LAN Emulation address
resolution protocol (LE-ARP) 
to gain the information it needs to associate a LAN destination with the
ATM address of another client.  
The LE-ARP response also contains information indicating whether a
destination device is MPOA capable 
or not. If the destination device is MPOA capable, the sending MPOA
client will monitor the amount of 
traffic it's sending to the destination.  If the volume of the traffic
reaches a pre-defined threshold, the 
sending MPOA will initiate an NHRP request in order to set up a direct
virtual circuit. By tracking the 
number of frames to a destination over a predetermined timeframe, the
client avoids the overhead of setting 
up shortcuts for short-lived traffic.

A number of vendors, including Fore Systems, Cisco Systems, Madge
Networks, and IBM, have indicated 
that they plan to implement MPOA.  Fore has a prototype of MPOA
implemented at a customer site already 
and expects to offer customers a full implementation in its ForeThought
5.0 release.  Cisco plans to deliver 
MPOA support in the next major release of its Internetwork Operating
System (IOS).  Madge will deliver 
MPOA support in the second half of 1997.

4.2 Fast IP

Fast IP provides a shortcut routing solution for packet only networks as
well as a solution that will work for 
both packet and cell based networks. Despite its name, Fast IP is not
IP-specific and can be extended to 
other protocols.  Like MPOA, Fast IP will enable two systems on
different VLANs to communicate 
directly, without going through a router, once the initial NHRP
request-response is completed.  Unlike 
MPOA, which is an ATM specific solution, Fast IP will work with FDDI,
Ethernet, token ring and ATM.  
NHRP, which was primarily designed for non broadcast networks like ATM,
has been changed to work on 
broadcast networks.

Both MPOA and Fast IP are based on NHRP. However in Fast IP when end
systems or edge switches issue 
an NHRP request, the request will include the originating system's MAC
address as well as its VLAN ID in 
the request packet.  When the destination system receives it, it will
issue an NHRP response directly to the 
originating system using the originating system's MAC address and VLAN
ID found in the NHRP request.  
Switches along the path of the NHRP response will forward the packet
based on either the originating 
system's MAC address or VLAN ID.

In returning the NHRP response, switches in the data path will learn and
map the address of the destination 
system.  Like the NHRP, once the originating system receives an NHRP
response, it will redirect data 
packets directly to the destination system using its MAC address, thus
bypassing the router.

In Fast IP, the intra-subnet communication stays the same.  The
inter-subnet communication out the same, 
using the normal controlled layer 3 path. However, once the controlled
layer 2 communication is 
established, Fast IP desktops and servers investigate whether there is a
faster, lower latency, layer 2 path 
available. A distributed Next Hop Resolution Protocol (dNHRP), that
unlike MPOA requires no dedicated 
NHRP server, is supported by peer Fast IP desktops and servers over the
controlled layer 3 path.  If both 
end systems support Fast IP and an end-to-end layer 2 path is
discovered, the desktop to server 
communication automatically moves over to the faster lower-latency layer
2 path.  If no layer 2 path is 
discovered, communication continues undisturbed over the original layer
3 path.

5 Comparative Analysis

Both MPOA and Fast IP discussed above attempt to address one or more of
the shortcomings of traditional 
routing.  Let's look at the similarities and differences of these two
approaches.

In terms of similarities, both MPOA and Fast IP are flow based schemes,
i.e. a shortcut is set up based on 
the volume of traffic going to a particular end station. Both MPOA and
Fast IP have value in the campus 
LAN environment, and where MPOA will appeal to customers with ATM
backbones, Fast IP will appeal to 
customers with pure packet-based networks. Because it's based on NHRP, a
key component of MPOA, 
Fast IP shares some of the same characteristics as MPOA. For example,
Fast IP inherently supports VLANs 
and is multiprotocol. Unlike MPOA, however, Fast IP can be deployed with
legacy packet-based 
technologies such as Ethernet and token ring. The fact that Fast IP is
implemented on packet-based 
hardware is a plus, since some customers may feel more comfortable
maintaining a pure packet-based 
network. However, the ATM camp is more bullish on connection-oriented
switching and featureslike traffic 
and congestion control, associated with ATM. Although ATM has not turned
out to be the "one-size-fits-
all" technology that some envisioned, it is a credible alternative to
traditional router-based networks. For 
corporate customers who have deployed ATM backbones, MPOA has clear
benefits. The most obvious of 
these benefits is the ability to communicate between subnetworks using
switched connections rather than 
the traditional hop-by-hop, packet-by-packet router scheme.


5.1 Quality of Service

MPOA version 1.0 does not have specifically-defined QoS support. LANE
version 2.0 does have 
mechanisms for associating QoS with LANE connections, and MPOA
implementations built on LANE v2 
will "inherit" this capability. The MPOA working group expects to
address the specific tie-in between 
MPOA and the IP-based Resource Reservation Protocol (RSVP) in the
future.

Similarly Fast IP has no explicit QoS support.  However, 3Com claims
that with Fast IP it has laid the 
foundation for future support for polices such as QoS.  The claim is
that functionality should not be added 
to the devices at the network edge and data centers at the network core
to let them monitor traffic flows 
across the network and make educated guesses based on those traffic
flows about what type of service is 
needed.  Instead, desktops and servers should be equipped with the means
to tell the network what they 
need and when they need it and then explicitly tag the associated
frames.  This will allow networks to 
implement required policies without adding the complexity of guessing or
compromising performance by 
having to carefully examine details in frames.  Fast IP involves
desktops and servers to improve the speed 
and lower the latency of networks by reducing layer 3 routing hops
whenever possible. 3Com claims that 
defining an active role for desktops and servers, as well as
streamlining the communication paths through 
the infrastructure, forms a powerful foundation for adding support for
policies such QoS guarantees.  
Combined with WAN specific schemes like Cascade Corporation's IP
Navigator, Fast IP can be extended 
to provide complete end to end support for QoS.

5.2 Price / Performance Benefit

As with all technology decisions, network managers and planners need to
weigh the benefits of shortcut 
routing against the costs. Do these shortcut routing schemes offer cost
or feature benefits that make them 
more attractive than next-generation multi-gigabit routers?

High-performance routers are currently being developed by various
organizations, including Cisco, start-up 
companies such as Juniper Networks, and some research organizations. For
example, BBN Research is 
building a 50 gigabit-per-second (Gbps) router that it expects to
deliver to the Defense Advanced Research 
Projects Agency this summer. These high-end routers should solve the
congestion problems posed by 
today's lower-performance routers, and could obviate the need for
shortcut routing schemes like MPOA 
and Fast IP. However, such high-performance routers are likely to cost
substantially more per port than 
shortcut routing alternatives such as Fast IP. If shortcut routing
alternatives continue to be more price 
competitive than standard routers, there will be a market for them.
Likewise these shortcut routing schemes 
offer other benefits that may make them more attractive than standard
routers, regardless of their 
performance. Both MPOA and Fast IP can simplify inter-VLAN
communications at the campus LAN level 
and reduce the role of routers in traffic forwarding, which should
translate into a need for fewer router 
ports. 

5.3 Migration Path / Standards

In a traditional collapsed backbone based on routers, MPOA doesn't offer
significant performance benefits 
unless you replace your existing routers with ATM switches. On the other
hand, if you've already migrated 
to a backbone based on ATM switches, MPOA offers significant benefits.
For organizations that are 
evolving their campus networks to include a combination of hubs, LAN
switches, and routers, Fast IP is 
worth investigating, particularly if you already use 3Com network
interface cards and switches.

MPOA requires changes only at the edge but not at the core of the
network backbone. MPOA works by 
moving traffic from an edge device at the entrance to the backbone to
the right exit device, rather than 
making traffic cross the backbone in a hop-by-hop fashion from router to
router. Similarly in Fast IP, since 
the NHRP protocol is supported by Fast IP desktops and servers, no
changes to switches and routers are 
required.  Only the Fast IP software needs to be installed on the
desktops and servers.  Desktops and servers 
with 3Com's Fast IP software will work with other vendor's switches and
routers.


Also the risk involved in deploying a given solution should be taken
into consideration. Although the 
specification is just nearing completion, MPOA can be considered
moderately stable in that all the leading 
ATM vendors have LAN emulation (LANE) implementations, and the upgrades
to LANE specified in 
MPOA aren't major. In addition, some vendors, such as Cisco and Fore,
already have implemented the 
Next Hop Resolution Protocol (NHRP), the other major component of MPOA.
Fast IP's stability is 
unknown since there is no wide scale deployment at this point.

Organizations evaluating shortcut routing schemes also need to decide
how important standards are to 
them. In general, a standards-based approach offers network managers the
best investment protection, since 
multiple vendors typically support any given standard. Network managers
therefore have the flexibility to 
buy standards-based products from more than one vendor with reasonable
assurance they will interoperate. 
However, due to the explosion of new standards, network managers may
find that vendors choose to 
implement different subsets of standards and that interoperability is
compromised.

MPOA is clearly standards-based, with a straw ballot by ATM Forum
members due in early 1997. 3Com's 
Fast IP combines three standard technologies, including a core protocol
used in MPOA. However, there is 
no other vendor that has committed to using the same set of standards in
quite the same way as 3Com. 
Officials at 3Com and IBM have indicated that they will provide
interoperability between their LAN 
switches based on at least one of the standards encompassed under Fast
IP, but it's unclear what degree of 
interoperability will be provided.

5.4 Protocol Support

MPOA can be viewed as a multilayer switching solution because it
encompasses both bridging and routing 
functions. As a result, MPOA supports both routable and non-routable
protocols such as IP, IPX, DECnet, 
AppleTalk, NetBIOS, and SNA/DLC, without needing to encapsulate them in
IP. MPOA allows subnets to 
extend across an ATM backbone, enabling moves and changes to be carried
out without the need for new 
IP addresses to be assigned.  Similarly Fast IP is not IP-specific and
can be extended to other protocols.

5.5 Scalability

MPOA relies on standard ATM signaling for connections to be established.
In addition, address resolution 
via NHRP must occur before an MPOA-based shortcut can be set up and
traffic begin to flow across the 
backbone. So MPOA is theoretically constrained by the fact that one VC
must be set up per shortcut 
connection. Thus many vendors characterize MPOA as campus-level
solutions due to concerns about its 
ability to scale to very large networks.  However, several vendors,
including Fore Systems and General 
DataComm, believe that MPOA can scale to the wide area. For example one
solution that gets around the 
VC problem is to have multiple flows aggregated onto a single connection
thus conserving VC resources. 

Like MPOA, Fast IP is designed for building and campus backbone.  To
extend Fast IP benefits to Wide 
Area Networks, 3Com suggests synergy between Fast IP and other schemes
like Cascade Corporation's IP 
Naviagor which address wide area scalability. IP Navigator shares the
Fast IP objective of creating fast 
paths by removing layer 3 routing hops across the network, while
maintaining layer 3 controls. IP 
Navigator addresses the VC scaling issue at layer 1, defining a
multipoint-to-point (MPT) tree VC for each 
site, significantly reducing the number of layer 3 routing hops across
the WAN. Therefore, by combining 
Fast IP and IP Navigator, it should be possible to maintain the control
of routing while capturing much 
more of the speed and low latency of layer 2 and layer 1 switching from
the desktop all the way across the 
WAN to a server at the other end.

The debate about which solution is the most scalable will only be
settled when proponents of each approach 
deploy them in WAN applications. Such deployments could begin this year,
but it will be 1998 before a 
clearer picture begins to emerge.

5.6 Filtering / Firewalls

Because shortcut routing approaches attempt to limit or avoid
packet-by-packet processing, network 
planners should be aware of how these schemes may affect their use of
routing filters for traffic control.
Filtering requires a router or other layer-3 device to examine the
layer-3 header of a packet. With MPOA, 
there is no way to "police" a VC once it's been established.
Consequently, any filtering must be handled at 
the edge of the network. For example, an MPOA server (by definition a
router) can refuse to forward an 
NHRP query if the sending station isn't allowed to talk to a particular
destination.
Similarly in Fast IP inter-subnet communication between two subnets
starts out the same, using the normal, 
controlled layer 3 path.  Thus, desired broadcast domain containment and
firewall safeguards are preserved 
across the network.

6 Conclusions

Both MPOA and Fast IP provide an easy way to optimize inter-VLAN
routing. While MPOA is an ATM 
specific solution, Fast IP also provides a solution for packet switched
networks which allows it to work 
with FDDI, Ethernet, token ring, and ATM.  Both MPOA and Fast IP have
value in the campus LAN 
environment.  MPOA will appeal to customers with ATM backbones and Fast
IP will appeal to customers 
with pure packet based networks. Both MPOA and Fast IP are flow based
schemes, i.e. a shortcut is set up 
based on the volume of traffic going to a particular end station.
There are some concerns about the 
scalability of these schemes to Wide Area Networks that haven't been
fully addressed yet.  Before adopting 
any of these schemes network managers have to look at a number of
factors including support for virtual 
LANs, quality of service support, protocol support, traffic management,
LAN-WAN integration and how 
these schemes meet future network design requirements.
MPOA and Fast IP are two of the many schemes being developed for solving
the router bottleneck 
problem.  It will be some time before these schemes stabilize and
industry has enough experience to 
determine which is most viable. Since the industry is in such a volatile
state, it's also possible that new 
routing alternatives will emerge that supersede the current crop of
solutions. Alternately, several of these 
approaches could be collapsed into one hybrid approach.

7 References

1. Multiprotocol Sub-Working Group: "Multiprotocol Over ATM Version 1.0
- Straw Ballot"  
http://www.cisco.com/public/rfc/ATM/atm-forum/mpoa/
2. Petrosky, Mary: "The Burton Group Network Strategy Report. Shortcut
Routing."
3. Hart, John: "Fast IP: The Foundation for 3D Networking,"
http://www.3com.com/FastIP/501312.html
4. Andres, Eric: "MPOA Ties It All Together,"  
http://www.data.com/Tutorials/MPOA_Ties_It_All_Together.html