Radio broadcasting, global concerts, and special events over the Net
are obvious examples of how the Internet is accommodating all-to-many
distribution of audio and video streams. While conspicuous � and increasingly
popular (at least 3,000 radio stations have already started broadcasting
over the Net) � these examples offer the merest hint of the possibilities
ahead, the potential for information sharing and dissemination that may
be realized once the techniques of one-to-many distribution are refined.
FORCE YIELDS TO FINESSE
At present, one-to-many distribution is usually accomplished by brute
force, that is, by networking techniques that replicate information at
its source and transmit the information to each receiver. IT managers
are wary of these types of applications, fearing that they�ll prove disruptive
� to an intranet, for example � because of the potentially huge demands
that they can place on network and server resources.
The potential for such disruption may be lessened or even eliminated
by something called IP multicast, a means of efficiently implementing
one-to-many communications. IP multicast avoids the processing overheads
associated with replication at the source and the bandwidth overheads
of repeatedly sending the same information to different destinations.
An analogy helps here. Like broadcast TV or radio, IP multicast establishes
a number of channels available across the IP network to which users can
�tune in.� In radio and TV, the signal is there whether you�re listening
or not. This is also true for any packet sent on a LAN, which is a broadcast
medium by its very nature.
With IP multicast, however, mechanisms are defined such that the network
replicates the content as close to the listeners as possible. In this
way, IP multicast avoids wasting bandwidth in those parts of the network
that lack listeners. Let�s look at how IP multicast is accomplished, and
then explore some of the applications enabled through IP multicast.
THE TECHNOLOGY BEHIND MULTICAST
Multicasting provides mechanisms to conserve network resources and minimize
server processing overheads by using intermediate nodes (for example,
routers and routing switches) to replicate data at the most efficient
downstream point(s) in the network. These mechanisms rely on two protocol
building blocks:
- A protocol that end stations use to join one or more multicast groups.
This protocol serves, in essence, as a tuning dial for end users, who
may signal their preferred channel.
- A protocol for establishing the optimal replication points in the
network given the location of the sending server and the listening clients.
Before discussing the protocols in more detail, we need to establish
a few basic concepts. �Multicast groups� define one or more transmitters
and a set of listening receivers. The construction of a multicast group
begins with the deployment of source video, audio, financial, or various
multimedia servers that will stream data into the network.
Another concept is the �multicast address.� Such an address is assigned
to the application for the duration of the multicast session. Receivers
signal to their local multicast-capable router their desire to join or
leave a multicast group. Alternatively, the router can poll the receivers
to see if they are still listening. When there are no receivers in a region
of the network, multicast packets will no longer be sent there.
These concepts are easy to understand, but how does the network establish
the optimal replication points given the location of the sending server
and the listening clients? At this point, we need to introduce a more
challenging concept, the �delivery tree.�
To establish optimal replication points, routers communicate with each
other via a network-layer multicast routing protocol that enables the
construction of a multicast delivery tree. The delivery tree is essentially
a set of paths calculated so those multicast packets are delivered only
to those network regions that require them.
Source-Rooted Trees
The dominant form of multicast delivery relies on source-rooted trees.
Each source within a group has its own tree that connects it to all the
receiving members of its group. If a group has five sources, it will have
five distinct delivery trees. The source-rooted tree takes the direct,
or shortest, path from the source to its receivers. Different routing
metrics can be used to compute the shortest path (hops, delay, cost, etc.).
Typically, the shortest path entails the fewest of hops between a source
and its receivers.
First-Generation Protocols: The first generation of widely used
multicast protocols was based on least-hop routing algorithms, specifically,
the algorithms used by the Routing Internet Protocol (RIP). When a source
sends its first multicast packet to the group, routers broadcast the initial
multicast packet to all interfaces except the one that leads back to the
source via the shortest path. This allows multicast frames to reach all
potential receivers.
During the flooding process, a multicast packet may be addressed to a
group that does not reside on any of the interfaces on a particular router,
with the result that prune messages are sent back towards the source.
These messages indicate that multicast packets should not be forwarded
down this branch of the tree. Trees may grow and shrink as receivers tune
in and out.
One advantage of this approach is that it places modest processing demands
on routers. Another is that it can be tunneled over a non-multicast IP
network, this being very attractive in the Internet as a means to kick-start
IP multicast networking.
Second-Generation Protocols: As the IP multicast market matures,
a second-generation multicast protocol has emerged. The new protocol overcomes
the limited convergence performance and scalability of first-generation
systems. It does so by eliminating the need to periodically flood multicast
traffic throughout the network. This protocol builds on the Open Shortest
Path First (OSPF) routing protocol that is the widely deployed successor
of RIP.
OSPF uses efficient link-state algorithms: instead of periodically exchanging
the number of hops to every router in the network, OSPF only requires
changes in the network to be exchanged. When a new group member joins
a multicast group, a special group membership message is propagated to
all other routers within a routing area, which add this information to
their link-state databases.
In this way, routers converge to create a detailed map of the multicast
topology. This sophisticated link-state approach allows the multicast
distribution system to adapt rapidly as group membership and network resources
change. If a physical network link goes down, the change is propagated
to each router�s database. The result is an immediate change in the multicast
route calculation. This approach can make use of flexible path calculation
metrics for source-rooted path tree construction, going beyond simple
hop count.
Shared Trees
While the above approaches use source-rooted trees, an alternative,
using what are called shared trees, has been developed for applications
that involve sparsely populated multicast groups with low-grade network
connections. An example of such a group is a video conference or video
broadcast that connects users in many dispersed locations that lack high-bandwidth
connections to the corporate backbone. Shared trees create multicast forwarding
paths that rely on a central router that serves as a rendezvous point
between multicast sources and destinations.
A shared tree has the potential to lessen demand on routers and network
bandwidth during tree construction. Although the shared tree avoids wholesale
flooding, it subjects multicast traffic to a static, non-optimized set
of paths that all pass through the rendezvous point router � a potential
bottleneck and single point of failure. To reduce the inefficiencies of
the shared tree, receivers or routers have the option to switch to a source-rooted
shortest path tree once the source starts multicasting.
Generic Router Assist
In certain application environments (for example, in financial networks),
a protocol is required for ensuring lossless, ordered, and duplicate-free
multicasting of critical data. Such tasks are ill served by TCP. This
protocol works very well in a point-to-point environment, but not in a
multicast environment.
More suitable protocols, including several reliable multicast Layer 4
protocols, have been developed. One example is Starburst�s Multicast File
Transfer Protocol. Another possibility, one allowing independent evolution
of the end-to-end Layer 4 protocol, shows promise. This protocol, called
Generic Router Assist, is designed to provide more effective recovery
from loss.
Generic Router Assist uses a negative acknowledgement-based mechanism
such that once a receiver determines that data has been lost, it attempts
to recover the data from sources close to the receiver. It goes back to
the source only as a last resort. For example, data may be recovered from
a router that may have cached the data or from another receiver that may
volunteer to retransmit the data.
THE BUSINESS OF MULTICAST
Multicast-capable network backbones are an essential component of emerging
multimedia, information distribution, and real-time computing applications.
Key drivers in the enterprise market include: distribution of inventory,
pricing, and stock market data; telemedicine; employee training; corporate
communications; new push applications; and the general area of multimedia
collaboration.
Multicast collaboration can increase the effectiveness of distributed
workgroups by allowing joint viewing and editing of common documents.
Such documents may include audio, image, and video components. Multicast
collaboration can also enable rich multimedia interaction between corporate
users, customers, suppliers, and partners over the Web. Online conferences,
discussion groups, virtual white board sessions, and shared Web applications
can be used to strengthen ties between a company and its customers.
In addition, IP multicast-enabled dynamic information distribution systems
will deliver more effective Web replication and caching, remote server
synchronization, and desktop software distribution. By providing efficient
ways of getting lots of information to lots of users, multicast-based
applications can give enterprises a powerful competitive advantage.
To further illustrate application enablement through multicasting technologies,
consider opportunities for:
- Specialized training � allowing expert coaches and instructors to
train athletes, performers, and artists in subtle motor skills even
though teacher and pupil may be separated by thousands of miles.
- Real-time 3D modeling � uniting groups of dispersed engineers and
scientists, who may view highly detailed graphical simulations of dynamic
scientific and engineering processes without leaving their desks.
- Editing of video footage and high-resolution animation � linking
artists and creative workers in virtual production studios far from
major media centers.
- Networked games � allowing players to compete with each other in
real-time over the Net.
CHOOSING A MULTICAST STRATEGY
To respond to business needs for multicasting applications, IT managers
will have to assemble adequate IP networking infrastructures. One thing
IT managers can do is implement policy-enabled QoS and access control
mechanisms. Such measures will help manage the unpredictable demands that
may assail a network. However, a strategy to deploy an effective infrastructure
involves more than just turning on a software feature.
For example, IT managers will have to tend to architectural issues.
Specifically, routers that are based on a central processing architecture
experience a significant drop in throughput when required to replicate
multicast packets. High-performance multicast demands distributed routing
switch architectures that can support multiple multicast groups and the
required replication function.
On top of this, real world considerations drive the need for multicast
migration tools. More generally, management tools are required to allow
introduction of multicast in an orderly fashion. Specifically, implementing
sophisticated filtering techniques may allow IT managers to stage the
rollout of multicast capabilities across an enterprise IP network.
In addition, IT managers might familiarize themselves with multicast
table management capabilities, accessing debugging tools capable of tracing
multicast operation as part of problem resolution. Another possibility:
multicast network tools enabling the visualization of multicast trees
for engineering and management purposes. Taken together, tools such as
these may facilitate multicast deployment in the enterprise.
MULTICAST FORECAST
IP multicast standards have been developed by the IETF. High-performance
multicast networking products are there and working in, for example, stock
exchanges around the world. The migration and management tools are there.
And the applications are emerging, enabling new forms of information sharing
and dissemination.
Tony Rybczynski is director of strategic marketing and technologies
for Nortel Networks� Enterprise Solutions unit. This business unit offers
a full range of enterprise terminal, workgroup, campus, and wide-area
unified networks and applications, through direct and indirect channels.
For more information, visit the company�s Web site at www.nortelnetworks.com.
E-mail questions or comments to tony.ryb@nortelnetworks.com.
|
TMCnet LOGIN
Webinars
