Multicasting: Architectures, Algorithms and Applications
May 2 - 4, 2001
DIMACS Center, Rutgers University, Piscataway, NJ
Presented under the auspices of the Special Year on Next Generation Networks Technologies and Applications.
- Micah Adler, University of Massachusetts, firstname.lastname@example.org
- D. Frank Hsu, Fordham University, email@example.com
Compression Using Efficient Multicasting
Micah Adler, University of Massachusetts
Many multiprocessor systems have the ability to broadcast
and/or multicast information efficiently. However, this ability is
often overlooked when designing algorithms for these systems. In this
talk, we introduce a new compression technique that uses efficient
multicasting to significantly reduce the amount of information
communicated during parallel and distributed computation, resulting in
significantly faster algorithms for Fast Fourier
Transforms and sorting on shared memory parallel models with limited
bandwidth. These algorithms demonstrate the importance of taking
advantage of efficient multicasting. The compression technique uses a
new, natural variant of Ramsey theory, which may be of independent
(Joint work with Tom Leighton)
Multicast in Optical Networks
Motivated by wavelength division multiplexing in all-optical networks,
we consider the problem of finding a set of paths from a fixed source
to a multiset of destinations, which can be coloured by the fewest
number of colours so that paths of the same colour do not share an arc.
We prove that this minimum number of colours is equal to the maximum
number of paths that share one arc, minimized over all sets of paths
from the source to the destinations. We do this by modeling the problems
as network flows in two different networks and relating the structure of
their minimum cuts. The problem can be efficiently solved (paths found
and coloured) using network flow techniques.
(this lecture covers work by B. Beauquier, P.Hell and S. Perennes)
Fault-Tolerant and Time-Step Optimal Multicasting
in Mesh Multicomputers Using Faulty-Block-Information
Dr. Xiao Chen
We propose a fault-tolerant and time-step optimal multicast algorithm
for mesh multicomputers based on the concept of the extended safety
level which is a vector associated with each node to capture fault
information in the neighborhood. In the approach each destination is
reached through a minimum number of hops. In order to minimize the total
number of traffic steps, three heuristic strategies are proposed. This
approach can be easily implemented by pipelined circuit switching (PCS).
A simulation study is conducted to measure the total number of traffic
steps under different strategies. Our approach is the first attempt to
address the fault-tolerant and time-step optimal multicast problem in
mesh muticomputers based on limited global information with a simple
model and succinct information.
Supporting Fast Intra-Domain Handoffs and Paging
using Multicast in the Dynamic Mobility Agent
Subir Das, Archan Misra, Anthony McAuley and
Sajal K. Das (Speaker)
In this talk we will first propose a new architecture, called
Dynamic Mobility Agent (DMA) architecture, for third and fourth
generation (3/4G) wireless cellular networks. The objective is to
reduce the latency of intra-domain location updates and the mobility
related signaling traffic. Then we will discuss the base protocol that
provide fast intra-domain handoffs by using a duration-limited,
proactive packet `multicasting' scheme. We will quantify the
expected buffering requirement of the proposed multicasting scheme for
typical 3/4G network characteristics and compare it with alternative
IP-based fast handoff solutions. We will also present a paging scheme
under the base protocol that replicates the current cellular paging
structure. Our paging mechanism supports generic paging
strategies and can significantly reduce the mobility-related IP
Our scheme allows the use of either mobile node (MN) initiated or
base station (BS) initiated handoff triggers and is applicable to
multiple future link-layer technologies. Moreover, the IP-based
fast handoff technique provides a secure fast handoff solution that does
not assume the existence of specific layer-2 authentication functions
or require the adjacent BSs to be aware of each other's identity.
Finally, the IP-layer paging mechanism, which allows an idle MN to be
located even though it does not perform IP-layer
at every change in the subnet (BS), makes the scheme relatively
independent of the radio technology.
Reducing the Multicast Routing State
IP multicast suffers from scalability problems for large numbers of
multicast groups. In mulitcast routing requires from each router to
keep forwarding state for every multicast tree passing through
it. Therefore, the state grows with the number of multicast groups.
In this work, we propose an approach to reduce multicast forwarding
states. In our approach, multiple groups are forced to share a single
delivery tree. We discuss the advantages and the imlementations
issues of our approach, and conclude that it is feasible and
promising. We then propose metrics to quantify state reduction and
analyze the bounds on state reduction of our approach. Finally, we use
simulations to verify our analytical bounds and quantify the state
reduction. In parallel, we study properties of multicast trees and
the behavior of the members. The initial simulations suggest that our
method can aggregate the routing state significantly.
Issues in Scalable Multicast Protocols
Art Farley, Virginia Lo,(Speaker) Andrzej Proskurowski,(Speaker)
We investigate two related ideas supporting development of scalable
multicast protocols. They concern new network substructures
(generalizing the standard distribution trees) and multiple
"dissemination centers" ("multiple cores"). Our investigations include
both theoretical analysis and experimental evaluation.
In particular, we will present some or
all of the following (time-permitting):
- Definition, distributed protocols for construction, and
experimental evaluation of shared network substructures
(spanners and stanners).
- Theoretical analysis of shared multicast trees for fixed set of sources
- Experimental evaluation of multiple core trees and a distributed
protocol for core placement ("k-domination")
- Experimental evaluation of SSM proxies, an application-level protocol
that implements any-source multicast on top of a source-specific
Computing connectivity of Cayley networks
The connectivity of a network measures the number of disjoint paths
between any two nodes, or from one node to several other nodes in the network.
It also measures the fault tolerance of the network.
Cayley networks (or Cayley graphs) are constructed from finite groups.
They have many nice properties that make them excellent candidates for
interconnection networks for distributed and parallel computer systems.
In this talk, we show how to compute the connectivity of any given
Cayley network that is connected (i.e. has only one component).
Our algorithm depends only on local information, that is, the degree
of a node and a small neighborhood of the node within a certain distance.
The algorithm is deterministic and has a polynomial time in term of the
degree. Note the number of nodes in the network is generally exponential
in the degree, so our method is efficient in a strong sense.
Service Location Protocols and Architecture for ad-hoc Networks
Ad-hoc networks are infrastructureless networks formed on-the-fly by
devices with wireless communication capabilities. They have bandwidth and
device power limitations, as well as the need to deal with mobility. If
client - server interactions are to be present in such networks, a method
for performing service location is needed. When two nodes that are several
hops apart communicate, resources on all the intermediate nodes are
wasted, so it is better to find services that are "closer" to the clients.
Existing service location protocols do not seem suitable for mobile
wireless environments. An underlying, low-level architecture that
addresses network access issues could enable existing protocols to work in
an ad-hoc environment. We have evaluated distributed and centralized
architectures for service location and examined service availability and
message overhead performance using simulations.
Building Dynamic VPN Multicast Trees Using Active Networks
Yaser Haggag and Sampalli Srinivas
Virtual Private Networks (VPN) have emerged as an important
solution to security threats (e.g. eavesdropping, data manipulation,
repudiation) surrounding the use of public networks for private
communications. VPNs provide strong security by integrating a
set of security components such as authentication, encryption, access
control and session management to proof the identity of both
communicating parties, to hide the exchanged data, to only
restrict access to legitimate parties and to insure efficient
operations of the established secure sessions respectively.
Furthermore, VPNs have been proven to be a commercial competitor
to the security offered by leased lines hence allowing
geographically dispersed parties to communicate economically
with each other [18,19].
With the increasing popularity of one-to-many and many-to-many
applications and services being deployed over public networks,
many multicast protocols are being developed to solve issues
in group communications [1,2,3,4]. One major issue is
scalability, which is a resultant of more members joining the
broadcast session thus overwhelming the server causing a decline
in server performance, which in turn slows server data delivery
. Another issue is security, which is required to keep the
broadcast session hidden and unmodified by unauthenticated/unauthorized
recipients [6,7]. One approach to solving the issue of scalability
is to move the broadcast session closer to the members by caching
the broadcast data onto the nearest routers securely, thus improving
scalability and reducing the server loads [8,9].
In this paper we introduce a novel approach under development
called "deployable virtual private networks" (DVPNs) to help
build dynamic (i.e. on demand) secure multicast trees between
the broadcast sever and the cached routers. The novelty of
the approach comes from the use of active network technology
to deploy VPNs. Active network is a new networking technology
that inserts intelligence inside the network to offer on demand
processing and programming capabilities to network nodes and
packets traversing the network [12,13]. This allows routers to
perform computations up to the application layer, thus increasing
the flexibility of the network to speed up the deployment of
new protocols and applications. Active networks offer many benefit
in terms of secure code execution, access control, congestion control
and data caching .
According to , several architectures have emerged to study
the benefits offered by active networks. One architecture
called active packets, inserts the programmable code only in the
packets travelling the network. Another called active nodes, builds
the programmable code into the intermediate nodes of the network.
However, a hybrid version combing the programmability of the
active packet and node approach appears to be the most promising one.
This paper presents the preliminary results of an ongoing
research project within the Secure Active VPN Environment
(SAVE) project in active networks at Dalhousie University.
PLAN (Programming Language for Active Networks) has been
selected as the execution engine for developing active network
applications and protocols on our test bed since it has emerged
as an important hybrid architecture
combining security, flexibility and efficiency [11,14,15].
This section describes the design approach under development
to deploy VPNs for dynamic multicast trees. To build a secure
multicast tree a few assumptions are in place:
(a) Routers are active.
(b) Routers employ Public Key Infrastructure .
(c) Routes with statically installed DVPN services are
trusted by the organization deploying the VPN.
(d) VPN multicast trees are most likely to be small or medium
in member's size.
(e) Trusted router can be trusted to enforce multicast
In addition to the core services available to PLAN programs,
we have successfully implemented and statically installed five
DVPN services (i.e. functions). Our goal is to dynamically
install DVPN services to demonstrate on demand VPN functionality.
The following is an overview of the implemented services:
1. chuckToBlob: This service converts a chuck (a code hunk
that contains code, a function to execute and bindings
(e.g.. arguments such as data and destination)) into a blob
(a byte array) to be able to encrypt it later using the
encryptBlob service .
2. encryptBlob: This service encrypts (scrambles) a blob.
By this we hide any data sent (what ever type it may be)
along with the service(s) required by it to be evaluated.
3. decryptBlob: This service decrypts (unscrambles) a blob
to evaluate it.
4. blobDigest: This service produces a hash value  of
the encrypted blob for
the purpose of message integrity check performed by
5. verifyBlobDigest: This service checks the integrity
of the received encrypted blob, then decrypts it to be
We have created a network configuration to test the operation
of the above PLAN services by creating a tunnel between two
active routers. To experiment with multicast like applications,
we have created a client chat program using Java (an object orient
programming language)  to simulate a broadcast session over active
networks. When the client chat program is injected, the routers interpret
the PLAN code calling on the cryptographic services installed
statically and applying the passed data to them. The implemented
cryptographic mechanisms ensure the exchanged chat text will
remain private. This design provides so far two different levels of
security, data confidentiality and integrity. The client can
either select to call on the encryption and integrity services or on
the encryption service only.
FEATURES OF THE DESIGN
The goal of this design is to develop VPN services that can
be installed dynamically and can be terminated on demand to
provide strong security mechanisms to a multitude of unicast
and multicast applications. Current traditional VPN solutions are
not flexible enough to offer on demand tunnel creation and
termination and are not easy to integrate with many applications
due to their complex interface. DVPNs can offer varying levels of
security, flexibility and usability (such as client-to-client,
client-to-node and node-to-node communication). The flexibility
of the DVPN services allows for the selection between a strong
secure procedure or a performance tailored one. For example,
data exchanges and backups between a main office and its
branch is a trusted one, on the other hand it is performance
demanding. Therefore, requiring fast encryption processing and
fewer authentications. Another example, communication between
businesses and customers can be tailored towards more
authentication than performance to ensure legitimate exchanges.
Furthermore, active networks allow VPN services to be installed
dynamically in a resource-controlled environment. Thus extending
DVPN reachability to traverse untrusted intermediate routers and
networks eliminating interoperability issues suffered by traditional
 B. N. Levine, J.J. Garcia-Luna-Aceves, "A Comparison of Reliable
Multicast Protocols", Multimedia Systems, 1998.
 R. Canetti, P-C. Cheng, D. Pendarakis, J. R. Rao, P. Rohatgi, D. Saha,
"An Architecture for secure Internet Multicast", IETF Internet Draft
(work in progress), February 1999.
 W. Fenner, "Internet Group Management Protocol", Version 2, RFC 2236,
 M. Calderon, M. Sedano, A. Azcorra, C. Alonso, "Active Network Support
for Multicast Applications", IEEE Network, May/June 1998, pp. 46-52.
 B. Quinn, "IP Multicast Applications, Challenges and Solutions", IETF
Internet Draft (work in progress), November 1998.
 M. J. Moyer, J. R. Rao, P. Rohatgi, "A Survey of Security Issues in
Multicast Communications", IEEE Network, November/December 1999, pp. 12-23.
 R. Canetti, B. Pinkas, "A Taxonomy of Multicast Security Issues", IETF
Internet Draft (work in progress), April 1999.
 L. H. Lehman, S. J. Garland, D. L. Tennenhouse, "Active Reliable
Multicast", IEEE INFOCOM, 1998.
 U. Legedza, D. J. Wetherall, J. Guttag, "Improving the Performance of
Distributed Applications Using Active Networks", IEEE INFOCOM, 1998.
 D. Alexander, W. Arbaugh, A. Keromytis, J. Smith, "Security in Active
Networks", University of Pennsylvania, 1999.
 M. Hicks, P. Kakkar, J. T. Moore, C. A. Gunter, S. M. Nettles, "PALN:
A Packet Language for Active Networks", in International Conference on
Functional Language (ICFP '98), Baltimore, MD, 1998.
 D. L. Tennenhouse et al, "A Survey of Active Network Research", IEEE
Communications Magazine, January 1997, pp 80-86.
 K. Psounis, "Active Networks: Application, Security, Safety and
Architectures", IEEE Communications Survey, 1999, Vol. 2, No. 1.
 M. Hicks, P. Kakkar, J. T. Moore, C. A. Gunter, S. M. Nettles,
"Network Programming Using PLAN", University of Pennsylvania. DARPA.
 Official PALN web site:
 Sun Microsystems Java web site:
 M. Hicks, A. Keromytis, "A Secure PLAN", University of Pennsylvania.
 D. Kosiur, Building and Managing Virtual Private Networks, John Wiley
and Sons, Inc, USA, 1998.
 B. Schneier, Applied Cryptography, John Wiley and Sons, Inc, USA,
 M. Hicks, P. Kakkar, "The PLAN Tutorial", v. 3.1. DARPA.
Scheduled RMP and a Stock Market Application
Nicholas F. Maxemchuk
The reliable multicast protocol,RMP (a.k.a RBP, a.k.a the
Chang/Maxemchuk protocol) was implemented in 1983, and was
the first reliable multicast protocol.RMP has several
characteristics that made it a likely candidate for the stock market
application. In particular:
1. the receivers in RMP all place the messages from all of
the sources in the same sequence,
2. we can eventually guarantee that all of the receivers have
a message, and,
3. at a later time we can guarantee that all of the receivers
know that all of the other receiver have the message.
Using RMP, we created three international stock market systems,
with useful fairness properties:
1. A unified stock ticker of the transactions that are being
conducted on the various physical and electronic exchanges.
This system simultaneously delivers the same composite ticker to
all receivers, anywhere in the world.
2. A unified sequence of buy and sell offers that are delivered to
a single exchange or a collection of exchanges. This system
gives all of the traders the same fair access to an exchange,
independent of their relative distances to the exchange or
the delay and loss characteristics of the international network.
3. A distributed, electronic trading floor. This system has the
fairness attributes of the first two systems and also conducts
irrefutable, distributed trades. The original version of RMP
implemented a distributed database on a local area network.
There were a few tens of sources and receivers and a small number
of messages per minute.
The stock market applications have to accommodate tens of millions
of sources and receivers, transferring millions of messages per
second, internationally, in a hostile environment, here protocol
participants cannot be trusted to cooperate. Most of the differences
in requirements between the original RMP application and the stock
market applications were solved architecturally, by good engineering,
rather than by modifying the protocol.During this presentation,
I will describe the architecture of the stock market systems and
show how it provides the fairness and security that are required.
The high message throughput and the short delivery delays in the stock
market applications required a modification of the protocol.
We solved these problems by changing the protocol from an event
driven protocol to a scheduled protocol. The original version of
RMP is driven by the arrival of source messages.Each source message
is only explicitly acknowledged by one receiver, instead of
by each receiver, by using a combination of positive
acknowledgements, negative acknowledgements, and token passing.When
a receiver accepts the token, it implicitly acknowledges all of the
previously, explicitly acknowledged messages.The penalty for
reducing the number of acknowledgement messages is the delay until
we know that all of the receivers have received an acknowledged
a message. When there are N receivers, the delay is the time
required to acknowledge the next N-1 source message, and is
therefore, a function of both the number of receivers and
the arrival times of future source messages.
The modified version of RMP sends acknowledgements at scheduled
times,rather than being driven by the source messages.
In this scheduled protocol if there is no negative acknowledgement
by a specified time it is the same as a positive acknowledgment.
As a result, we can guarantee that all of the receivers have a
source message at a predetermined time after it is first acknowledged,
independent of the number of other receivers or the message arrival rate.
I am currently determining how tight I can make the interval between the
time a message is first transmitted until I can guarantee that all
of the receivers have the message.The time depends on the network
loss rate and the propagation delay. In certain networks the
guarantees are adequate to meet the quality of service required
by real time applications.
Hierarchical Path QoS on the QoS-based Multicast
Yasuo Okabe, Takaaki Sekiguchi, Kenji Fujikawa, Kazuo Iwama
We argue on a method to collect information of each
existing multicast flow on hierarchical networks, and describe an
algorithm for Qos multicast routing based on it.
In computing a multicast path with per-flow QoS guarantees precisely,
it is an important problem how routers collect information of
each existing multicast flow. SRSVP, a QoS-based multicast routing
is designed as it collects flow-specific detailed information,
called PathQoS, by putting it into signaling messages, so that
the derived QoS path becomes more efficient.
In this paper, we have designed an algorithm to calculate PathQoS
corresponding to an aggregated link state among areas on a hierarchical
network for SRSVP to compute more efficient QoS paths.
We have presented that the methods to calculate PathQoS among areas are
classified to four cases, have considered dependence upon information
needed by the calculations, and have designed an algorithm for routers
to be able to process the methods efficiently. We have implemented
the mechanism to collect PathQoS based on the algorithm.
Now SRSVP is fully adapted to a hierarchical network model.
Bid-based Pricing Approaches for Multicast Sessions in
Heterogeneous Networking Environments
Dan Rubenstein and Micah Adler
One reason that multicast has seen limited deployment is that ISPs are
unsure of how they should charge for such a service. A justifiable
pricing mechanism must satisfy three requirements. First, it must
guarantee the network a profit. Second, the protocols that implement
the pricing mechanism should must themselves be simple and efficient.
Last, the mechanism must permit admission of sessions that allow for
session heterogeneity: receivers within a session should be allowed to
receive at differing rates, and it should not be assumed that all
nodes within the network will provide multicast (i.e., some form of
multicast overlay may be required). In this talk, we demonstrate that
it is indeed possible to efficiently implement a bid-based pricing
mechanism in a multi-rate overlay multicast that optimizes network
profits when a session's transmission is restricted to a single tree.
We also explore the complexities that arise when we attempt to
extending the session's transmission beyond a single tree. Last, we
demonstrate how these bid-based protocols can be used in heterogeneous
environments to implement pricing mechanisms that are strategyproof,
such that receivers have no incentive to bid less than what receiving
the transmission is worth to them.
Distance Learning using IP Multicasting for the Ateneo
Rafael P. Saldana
The Ateneo Educational System consists of schools located in various
parts of the Philippine archipelago. Lead by the Ateneo de Manila
University, a distance education program is currently being implemented
in the educational system. Due to IP multicasting technology,
the delivery of high performance computing applications over the
the Internet is now a reality. To enhance the delivery of distance
learning in the Ateneo educational system we are proposing the
utilization of IP multicasting.
How to spread a rumor when not everyone is talking
to each other OR Multicasting in heterogeneous networks
Baruch M. Schieber
In heterogeneous networks, sending messages may incur different
delays on different links, and each node may have a different
switching time between messages. The well-studied telephone
model is obtained when all link delays and switching times are
equal to one unit. We investigate the problem of finding the
minimum time required to multicast a message from one source to
a subset of the nodes of size k. The problem is NP-hard even in
the basic telephone model. We present a polynomial-time algorithm that
approximates the minimum multicast time within a factor of O(log k).
Our algorithm improves on the best known approximation factor
for the telephone model by a factor of O(logn/loglogk), where n
is the total number of nodes. No approximation algorithms
were known for the general model considered by us.
This is joint work with Amotz Bar-Noy, Sudipto Guha and Seffi Naor
A Distributed QoS Multicast Routing Protocol
Many Internet multicast applications such as teleconferencing
and remote diagnosis have Quality-of-Service (QoS) requirements.
Such requirements can be additive (end-to-end delay),
multiplicative (loss rate) or with a bottleneck nature (bandwidth).
For these applications, QoS multicast routing protocols are
important in enabling new receivers to join a multicast group.
However, current routing protocols are either too restrictive in their
search for a feasible path between a new receiver and the multicast
tree, or burden the network with excessive overhead.
In this talk I'll present S-QMRP, a new Stateless QoS Multicast Routing
Protocol that supports all three QoS requirement types. S-QMRP is
scalable because it has very small communication overhead and
requires no state outside the multicast tree; yet, it retains
a high success probability. S-QMRP achieves the favorable tradeoff
between routing performance and overhead by carefully selecting the
network sub-graph in which it conducts the search for a path
that can support the QoS requirement, and by auto-tuning the
selection according to the current network conditions.
S-QMRP does not require any global network state to be maintained
and can operate on top of any unicast routing protocol.
Our extensive simulation shows that S-QMRP performs better than
the previously suggested protocols.
Joint work with Shigang Chen
Some Techniques for Multicasting in Hypercubes
We consider techniques to minimize message congestion when doing
multicasting in a hypercube architecture. That is, for any integer k, we
consider one-to-k maps which specify, for each node x in the hyper cube,
to which nodes packets are to be sent from x. Our results give techniques
to route such packets that satisfy the requests in specified sets of
one-to-k maps and that guarantee minimal message congestion in any of the
The Challenge of Multicasting over Ad Hoc Mobile
Multicasting over wireless networks has traditionally been investigated in
the view of mobile users moving from the radio coverage of one cell to
another. Current multicasting technqiues used in the Internet are robust
but not necessarily efficient when applied to wireless networks. Most
multicast routing schemes rely on building some forms of multicast trees,
be it source- or core-based. In addition, assumptions were made that
intermediate routers in the tree paths do not move. However, an emerging
new form of totally wireless and pervasive networks known as ad hoc
wireless networks have evolved. Such networks has a completely wireless
backbone, and use multiple wireless links for end-to-end communications.
Multicasting of audio/video/data information in such networks shall
present great challenges to researchers due to the presence of mobility,
bandwidth, and power constraints. One should question the suitability of
multicast trees in such networks and also suggest mechanisms to deal with
group membership dynamics. One should also examine scalability issues
associated with multicasting in an ad hoc based wireless intranet. This
talk shall reveal these problems and encourage thoughts for further
Deadlock-Free Prefix Multicasting in Irregular Networks
Jie Wu, Fei Dai, Li Sheng
A deadlock-free multicast scheme called prefix multicasting
in irregular networks is studied. Prefix routing is a special
type of routing where a compact routing table is associated with
each node (processor). Basically, each outgoing channel of a node
is assigned a special label and an outgoing channel is selected if
its label is a prefix of the label of the destination node. Node
and channel labeling in an irregular network are done by using
a pre-defined spanning tree which may or may not be minimum.
The routing process follows a two-phase process of going up and
then down along the spanning tree, with a possible cross channel
between two branches of the tree between two phases. It is shown
that the proposed routing scheme is deadlock- and livelock-free. The
approach is extended to multicasting in which the multicast packet is
first forwarded to the longest common prefix (LCP) of destinations in
the multicast. The packet is then treated as a multi-head worm that
can split at branches of the spanning tree as the packet is forwarded
down the tree. Comparison between prefix multicasting and
Libeskind-Hadas et al's approach for multicasting is given.
Possible extensions are also discussed including reliable prefix routing
A Scalable Topology for High-Speed LAN and WAN Networks
Sotirios G. Ziavras
We propose a topology to implement backbone networks in the LAN
or WAN environment. This network topology assumes the employment of
fiber optic devices. Wavelengths used in this scalable network are
reusable and a high degree of reliability is supported. We present
efficient algorithms for randomized many-to-all broadcasting and
multicasting in the proposed network. These implementations are tested
under realistic communication traffic patterns and message generations.
A modification of the multicast algorithm applies congestion control
to improve the performance.
Contacting the Center
Document last modified on May 1, 2001.