Ngi-07.dvi

A Survey of Reliable Multicast Communication Adrian Popescu, Doru Constantinescu, David Erman, Dragos Ilie Dept. of Telecommunication Systems Abstract— The paper reports on recent developments and
of the most important issues to be considered in reliable multi- challenges in reliable multicast communication, with special focus
cast communication. Section IV describes the main solutions on reliable multicast communication at the application layer.
existent today for reliable multicast communication. Finally, The foundation of reliable multicast communication is given
by several components, which are multicast communication,
congestion control and error control. Our paper is providing
a survey of these mechanisms in multicast environments.
Multicast communication represents the operation of send- Group communication has emerged as one of the most ing a packet to a group of recipients, which may be scattered important developments in Internet. Video conferencing, mul- throughout the Internet. A single SEND operation is used in timedia distribution, online gaming and long-distance educa- this case to deliver copies of packets to all receivers. The tion are today some of the most popular Internet applications, source address is a unicast address, whereas the destination which generate large amounts of traffic. To support these ap- address is a group address of some specific type.
plications, reliable multicast communication is a prerequisite.
Unlike broadcasting, multicasting allows every member to The purpose is to provide efficient and reliable communication choose whether to be part of the multicast group or not.
services among a number of users, who are members of a Multicasting is a way to reduce network load and end-to-end (e2e) delay. It can be used in conjunction with caching to Traditional multicast communication demands for the pres- improve the scalability and delivery performance. Multicasting ence of a multicast group, together with associated facilities is therefore most beneficial to users that source the information for reliable multicast communication, to which the users can as well as to ISPs and carriers. However, efficient multicast subscribe and participate. Even though IP multicasting was communication demands for special capabilities and specific introduced twenty years ago [14], it is still not widely available algorithms at various layers of the protocol stack. As a as an open Internet service. Problems related to per-group minimum, a multicast service should provide several basic state maintaining, scalability, reliability, congestion control and security have been postponing the wide deployment ofIP multicast.
On the other hand, other solutions have been developed for multicast service, to compensate for the above-mentioned limitations, e.g., MBone [18]. MBone provides an overlay network, which connects IP multicast capable islands by using The goal is to satisfy users, network operators and content unicast tunnel connections. Furthermore, other developments related to, e.g., video distribution and long-distance education,has further pushed the research and development of new alternative solutions for multicast, which are implemented atthe application layer.
Multicasting has been implemented at different layers in Our paper is a survey on current solutions for multicast the protocol stack, i.e., physical layer, network layer and communication as well as on solutions for the provision of application layer [20], [27], [32]. Today, some of the most reliable communication in this context. By reliable multicast communication we mean a type of multicast communication that has included facilities of error and congestion control.
The rest of the paper is as follows. Section II provides a Application layer (AL) multicast of type Peer-to-Peer survey of multicast communication. Section III presents some This work has been done within the research project ”Routing in Overlay AL multicast of type Overlay multicast (OM) Networks (ROVER)”, granted in 2006 by the EuroNGI NoE.
AL multicast of type Waypoint multicast (WM) 1) Physical layer multicast: A good practice used in mul- itself is relieved of these responsibilities, and only needs to ticast is that hosts receive and process only packets that are provide the basic stateless, unicast, best-effort delivery. The addressed to them. This can be done at the link layer. This is architecture therefore decouples multicasting from the unicast because every packet received by a network interface causes routing infrastructure, which gives important advantages in an interrupt in the device driver. This may generate further terms of ease of deployment and flexibility. However, AL processing at higher layers. A good solution could be in this multicast faces a number of challenges related to routing, case to use the so-called ”multicast filters”, to add multicast efficiency, reliability and scalability, which must be solved in facilities at network interfaces, and to distribute data locally in a LAN environment [32]. This solution is known as physical Over the last years, AL multicast has been the subject of layer multicast (PHY multicast). The performance of PHY much research and development, in spite of relative drawbacks multicast is however limited, especially due to the lack of like inherently being less efficient than IP multicast in using flexibility. A better solution could be in this case to use a network resources (packet duplication on unicast links cannot mapping between IP multicast or application layer multicast be eliminated), soft QoS guarantees and increased complexity of the end system [12], [20], [27]. Furthermore, by means 2) IP multicast: Multicast facilities can be provided at the of cross-layer communication, the overlay network can be IP level as well. The ”IP multicast” solution (fig. 1(a)) is a organized such as to provide topology-aware multicasting.
solution originally put forth by Steve Deering in 1989 [14].
Using various techniques, end hosts may collect information IP multicast provides support for both efficient group manage- from IP routers to build up more efficient AL multicast ment and efficient packet forwarding through the network. It is networks and to reduce packet duplication.
based on an open service model, which does not restrict users We distinguish three categories of AL multicast, namely to create or join multicast groups. Furthermore, senders are not Peer-to-Peer (P2P) multicast, Overlay Multicast (OM) and required to belong to a multicast group. The Internet Group Management Protocol (IGMP/IGMPv2/IGMPv3) is used in 4) P2P multicast: P2P multicast means that there are only connection with IP multicasting to allow a multicast router end hosts that handle the basic functions (group membership, to learn the addresses associated with networks attached to it addressing and routing), whereas in the case of OM multicast and to allow hosts to announce interest in receiving multicast there are a number of strategically deployed overlay proxy to edge routers [23]. The group management protocol is an nodes used to back-up the end hosts. P2P networking was orig- integral part of the IP layer in all hosts and routers that inally designed for information sharing and messaging (e.g., support multicasting. Furthermore, other important questions Napster, Gnutella) and it offers several important advantages are those regarding the multicast source type (e.g., Any-Source in terms of, e.g., self-scaling, which means that when more Multicast, Source Specific Multicast), multicast addressing and end hosts join the multicast group more bandwidth is supplied multicast routing (e.g., SBR, Steiner Tree, PIM) [12].
[20]. The price however is in terms of dependence on host IP multicast has important advantages that significantly bandwidth and loosely coupled relationships among the peers improve efficiency in content distribution. These are effective (with impact on QoS). Other important advantages are related when the physical media is broadcast in nature, and also to flexibility and lack of dependence on the unicast routing efficient utilization of link bandwidth and efficient content infrastructure. Furthermore, a specific challenge is in this case replication. Altogether, one can state that IP multicast is well the need to handle the presence of high churn rates in P2P suited for large-scale content distribution, especially for live, networks [15], [43]. An important consequence of high churn rates is that the topology is very dynamic, which makes it The openess of the model may however create serious difficult to provide hard QoS guarantees.
problems with the consequence that there is a real risk that 5) OM multicast: OM multicast has an alternative archi- the global deployment of IP multicast may be postponed tectural solution, which means that the content is distributed indefinitely [38]. Other important issues are related to the to proxy servers located close to end hosts. The group of need to support per group state in routers (with impact on proxy servers are organized into an overlay network to provide scalability), problems related to class D addresses (lack of delivery service to end hosts. Better performance can be of- hierarchy, limited number of addresses, long-term transition to fered here in terms of, e.g., maximized bandwidth, minimized IPv6), security problems and business-related problems (e.g., latency/jitter, improved accessibility [36], [37]. Akamai is lack of standards for charging of multicast services).
perhaps the best example of a Content Distribution Network 3) AL multicast: Another solution for multicast is to pro- (CDN) provider that is using this model for video streaming vide multicast communication at the application layer (so- delivery [1], [26]. Multicast networks are composed in this called AL multicast), and to use the unicast transport facilities case by multiple Points of Presence (PoP) with clusters (so- offered by TCP and UDP. Operations related to group mem- called Surrogate Servers) that maintain copies of identical bership, addressing and multicast routing are implemented content, thus providing better balance between cost for content in this case at the application layer on the end hosts of providers and QoS for customers. CDN nodes are deployed in a network. Application specific intelligence can be used to multiple locations, placed in different backbones all over the develop efficient multicast services. Consequently, the network world. They cooperate with each other, transparently moving A fundamental goal of the process of building the multicast group is to create a loop-free topology to serve, e.g., contentdistribution to members participating in the group. A logicaldistribution tree is constructed, which is rooted at the source.
Depending upon the relationship among nodes, they can bepartitioned into two main categories, i.e., parents and children.
The process of building the multicast group is a sophisti- cated one, especially in the case of AL multicast. There are a number of fundamental steps that must be considered insuch a process, i.e., peer discovery, neighbor selection, parentselection and group maintenance [16].
Several performance metrics have been defined to charac- terize the multicast communication service and impacts on thenetwork [17], [44]. The most important metrics are: Link stress, in terms of number of identical packets aphysical link carries.
Link stretch, also called relative delay penalty. This isthe ratio of delay between two nodes along the overlaydistribution topology to delay of the direct unicast path.
Resource usage, in terms of the sum of the delay ∗stressproduct over all participating links.
Time to first packet, which is the time required for a new member to start receiving data after joining a group.
Losses, which is the average number of packet lossesafter an ungraceful failure of a single participating node.
Robustness to changing network conditions.
The algorithms for topology creation can be implemented in different ways, each of them with different advantages and content to optimize the delivery process and to provide users the most current content. The optimization process may result, e.g., in reducing the bandwidth cost, improving availability and Centralized algorithms, with partial or full membership The client-server communication flow is replaced at OM Distributed, self-organizing algorithms, which differ inthe way the topology is created (e.g., mesh first, tree first) by two communication flows, namely one between the originserver and the surrogate server and the other between the sur- Desired features of such an algorithm are good perfor- rogate server and the client. Requests for content delivery are mance (not much worse than IP multicast), scalability, ease intelligently directed to nodes that are optimal with reference of deployment, robustness (respond well to changing network to some parameter of interest, e.g., minimum number of hops, conditions), quick and fair response to changing conditions of or networks, away from the requester. However, questions group membership and security. Today, most group construc- related to QoS provision, content multicasting and multipath tion algorithms seem to be distributed and self-organizing such routing heavily complicate the picture.
as to reduce the stress on the source node and to allow for goodscalability performance [20].
6) WM multicast: An alternative solution for AL multi- Another important algorithm, which takes over after the casting is given by the Waypoint Multicast (WM) solution, multicast group is constructed, is performance-aware adapta- as described in [10], [20]. Waypoint nodes are specific nodes tion of the e2e performance function to dynamics of multicast existent in a pool of common resources, which may be invoked group and changing network conditions.
to temporarily enter a multicast group and to provide the There are several strategies to construct AL multicast over- lacking bandwidth needed at the specific moment to support all multicast hosts. These nodes can be statically or dynamically provisioned. The behavior of a waypoint node is similar to that of an end host (as used in P2P multicast), the advantage however is given in this case by higher degree of flexibility 1) Mesh-based overlays: The mesh-based approach means at different rates, i.e., multirate video streams [13], [29]. Each that the nodes are organized in a mesh topology, where of these streams can be independently decoded and the content every node has knowledge of a set of other nodes, called reproduced with different QoS degrees, depending upon the neighbors. An important feature is that there is more than one path available for communication between an arbitrary Besides content replication, another useful concept is con- pair of nodes and neighbors are cooperating to exchange the tent decomposition. In such a case, a raw video sequence is content according to some predefined cooperation strategy.
compressed into several non-overlapping video streams (so- This means that alternative paths already exist without the called ”layers”), and dedicated tree/mesh topologies can be need to reconstruct the path between two nodes in case of used in the multiple tree/mesh overlay to carry the specific negative events, e.g., path crashes. Another positive feature is streams [13], [29]. The receiver can selectively subscribe to a that this offers advantages with respect to routing stability as number of layers based on the resources it has, e.g., in terms of available bandwidth. QoS can in this case be improved when The main drawback of the mesh-based approach is related to more streams are received and decoded together.
difficulties in constructing loop-free forwarding paths among There are two categories of layering schemes. These are the group members. Other drawbacks are the increase of link cumulative layering and the non-cumulative layering. Cumula- stress, complexity of algorithms needed for cooperation strat- tive layering means that one layer has the highest importance egy as well as for chunk selection strategy [2].
and contains the parts of content (e.g., video) with most 2) Tree-based overlays: The tree-based approach means important features. Additional layers are called enhancement that a specific algorithm is used to build up a tree topology layers and contain parts of content that progressively refine such that a single path is established between two arbitrary the quality of reconstructed content. On the other hand, non- nodes. Two of the most popular algorithms are the recursive cumulative layering means that all layers have equal impor- algorithm and the clustering algorithm [17]. In the case of the tance in content reconstruction and any set of layers can be recursive algorithm, a newcomer node first contacts the tree used for this purpose. The flexibility is therefore higher in the root, and selects then the best node among the children of the root node with respect to some reference set of parameters.
The multiple tree/mesh approach offers the advantage of This procedure is repeated until an appropriate parent is finally reducing the impact of network and group dynamics by using selected. The clustering algorithm first creates a hierarchy of decomposition and redundancy. The price is in terms of TCP clusters, and then newcomers recursively cross it to find the non-friendliness, scalability, and responsiveness. Intensive re- search is therefore done to solve these problems [29].
Some interesting tree-based architectures are [2], [6]: 4) Ring and multi-ring overlays: Another solution for Linear architecture, where clients are organized in a chain group communication is ring and multi-ring overlays. These architectures have significant advantages over mesh and tree Tree distribution with outdegree (k T reek), where clients overlays in terms of reliability, survivability and security [46].
are organized in a tree with an outdegree k and an interior Tree- and mesh-based architectures have inherent flow and node in the tree serves k clients simultaneously.
congestion control problems, especially in the case of using Forest of parallel trees (P T reek), where a specific con- the traditional ACK reliability-based error control [32]. On tent is first split into k parts, each part is then distributed the other hand, ring and multi-ring overlays have advantages over an independent tree rooted at the server, and finally in terms of inherent reliability and single fault tolerance. This the content is reconstructed at the receiver.
is because of the ring-based topology itself, where packets The tree-based approach is especially advantageous for one- are easily looped back to the sender. Another advantage is to-many multicast communication, which is typical for content given by the low number of ACKs needed in this case. There distribution networks. This means that, e.g., a content provider are even situations where no ACKs are needed to provide a first sends the content to the root node for further distribution successful communication. This is because the original packets to multicast nodes. This is the typical communication model are easily looped back to the sender in the case of successful used in IP multicast, although larger amounts of data can be transported in the case of AL multicast. Compared to IP Ring and multi-ring group communication have the draw- multicast, the AL multicast has the drawback of larger amount back of longer communication paths and, accordingly, larger of resources needed to provide the multicast communication delays and jitter. Furthermore, another possible drawback is service as well as the risk of inefficient use of available related to scalability, but this can be improved by building up hierarchical architectures of smaller multi-rings interconnected 3) Multiple tree/mesh overlays: The multiple tree/mesh together to replace large single rings [46].
approach represents an attempt to open up the bottlenecks of the above-described architectural solutions and to remove (DHTs) is an approach developed with the purpose to effi- so the limitations of the mesh- and tree-based approaches ciently construct a tree such as to solve the problem of receiver [20]. The fundamental concept is to use a specific codec that scalability and efficient location of data items [8], [39]. The generates replicated (video) streams for the same content, but fundamental concept is to develop a distributed infrastructure to provide hash-table functionality on Internet-like scales. A decentralized algorithm is used for this. Hash table semantics There are two possibilities for doing error correction.
are exposed in this case over a multicast group of nodes.
These are retransmission of corrupted data and the use of Every node may insert or retrieve a value associated with a parity packets in data, i.e., the so-called Forward Error key. Ideally, the keys and the associated values are uniformly distributed across all nodes. The exact distribution of keys and Where to perform error and congestion control in a values is highly dependent on the hashing function employed in the specific DHT. The basic operations of insertion, lookup There are two possibilities to do error and congestion and deletion of (key, value) pairs can be performed in a DHT control mechanisms in multicast communication. These network. A routing algorithm is also used to allow any node are hop-by-hop and end-to-end mechanisms.
to route to the node associated with a specific key.
What type of congestion control mechanism to use? DHTs have been shown to provide scalable routing and Depending upon the regulating parameter, there are four indexing, robustness and low latency properties. DHT is par- mechanisms for doing congestion control in multicast ticularly advantageous for large scale distribution networks, environments. These are the window-based, rate-based, e.g., simulation studies have shown latencies that are less than layer-based and local recovery-based mechanisms.
twice the IP path latency in case of networks with 260 000 How to do congestion control in a fair way to competing nodes [39]. An important drawback is however sensibility to How to develop scalable solutions for error and conges-tion control in multicast communication? III. ISSUES IN RELIABLE MULTICAST COMMUNICATION How to open up the performance bottleneck related Due to diverse and challenging conditions, reliable multicast to cachability and cache consistency, which limits the communication has been shown to be a difficult task [24].
Reliable multicast communication is requested to perform wellunder the conditions of heterogeneity of nodes (in terms of, Depending upon the specific situation and conditions for e.g., different processing capacities) and of the transmission multicast communication (e.g., IP multicast, AL multicast), channel (in terms of bandwidth, loss and delay characteris- different solutions can be used that are suitable for the specific tics), heterogeneity of content (static content, dynamic content case. Furthermore, another important parameter that influences and streaming media) with different characteristics and QoS the mechanisms developed for error and congestion control requirements in distribution, and also other specific conditions is related to the delivery service model used in case of (e.g., scalability, group dynamics and particular limitations in content distribution. There are three delivery service models the effectiveness of caching). Appropriate protocols should be designed for error and congestion control in a multicast com- munication scenario, which are able to provide the requested This is a synchronous service model, where a sender performance in terms of, e.g., error rate, delay and fairness to initiates concurrent delivery to all receivers in the group competing traffic flows on shared links.
and the receivers are supposed to be ready before the There are several issues that must be addressed by such transmission begins. The goal is to minimize the syn- chronization between the sender and the set of receivers.
Where to perform loss detection in a multicast commu- Various mechanisms for session announcements, session management and receiver reports can be used in combina- Loss detection can be done either at the sender or at the tion with this service model. The push model is particu- larly attractive for satellite and wireless communications.
There are two types of feedback messages that can This service model is particularly attractive for the dis- be used, namely positive acknowledgments (ACKs) and tribution of popular content. The content is continu- ously multicasted to receivers by using some specific distribution mechanism such that the receivers may join There are two alternative solutions possible in this case.
the group, download the content and leave the group These are either via unicast communication back to whenever they want. The performance is independent of sender or via multicast communication to all members loss patterns and session joining time. The service is scalable as well, although non real-time.
Who is responsible for retransmitting in case of corrupted This service model is typically used for delivering of In the case of multicast communication, data retransmis- audio and video content. Streaming objects are usually sion can be done from three different places. These are the much larger than Web objects and the consequence is sender, one of the receivers and one of the intermediate that timeliness is more important than the transmission nodes that has a copy of the original data.
reliability. Delay jitter between servers and clients is also more important than, e.g., compared to Web content feedback implosion when the feedback from all receivers may delivery. Furthermore, the streaming service does not overwhelm the sources and links close to source [28].
typically lend itself to caching, and the consequence is Another limitation of window-based regulation is related to that there is a need for closer cooperation between the fairness, i.e., the risk that other TCP sessions are driven into producers of content and the content delivery network.
The problem of reliable multicast communication refers to As mentioned above, content replication and content decom- both IP multicast communication and AL multicast commu- position can be used in combination with a multiple tree/mesh nication. Just like the case of a unicast communication that topology to provide multicast communication. In such a case, may require TCP on top of IP unicast, a multicast application a raw video sequence can be compressed into several non- may require a reliable multicast communication on top of overlapping video streams (so-called ”layers”), and dedicated IP multicast. Techniques similar to those used by TCP for tree/mesh topologies can be used in the multiple tree/mesh unicast communication can be used for multicast communi- overlay to carry the specific streams.
cation as well, e.g., window-based congestion control, use A particular feature of layer-based congestion control is that of sequence numbers, positive acknowledgments. There are this is a receiver-based approach. This means that it is the however significant differences, in the sense that mechanisms receiver that autonomously decides whether to subscribe to for reliable multicast communication should be able to handle, the multicast group or not. Based on the available resources, in a scalable manner, highly heterogeneous receivers and to the receiver may also decide on how many layers to subscribe cope with highly dynamic network conditions.
to or to drop. Each receiver should also detect packet lost on There are several important questions related to reliable and the way to it, and to adapt the window size or nominal rate. In scalable multicast communication, like for instance: some specific cases, the receiver should determine the RoundTrip Time (RTT) from the source as well.
What is the best place for controlling network congestion, Examples of implementations of layer-based congestion control are the Asynchronous Layered Coding (ALC) [24], What is the most suitable regulation parameter for mul- Receiver-driven Layered Control (RLC) [45] and Layered ticast communication, window-based or rate-based? Video Multicast Retransmission (LVMR) [30].
What is the best implementation for congestion and error In spite of some difficulties (e.g., TCP non-friendliness, control in AL multicast communication, hop-by-hop or scalability problems), the layer-based congestion control mechanism offers advantages with reference to scalability and The goal of a reliable multicast communication is to de- the heterogeneity that may exist in a multicast group, e.g., in sign scalable mechanisms for congestion and error control terms of network bandwidth. Depending upon the available in multicast environments with similar behavior as TCP, and local resources, a client may subscribe to a particular number to allow fairness in resource sharing. Some of the most of layers irrespective of the other clients. This mechanism is popular congestion control mechanisms used in multicast also advantageous with reference to the heterogeneity in user communication are window-based congestion control, layer- behavior and to solving the fundamental conflict existent in a based congestion control, rate-based congestion control, and multicast group between the asynchronous behavior of users local recovery based protocols [3], [12], [16], [24], [32].
and the synchronous nature of multicast communication.
A. Window-based congestion control The window-based regulation has three important limita- The rate-based regulation is, in principle, a mechanism tions with impact on scalability. One of them is given by the that keeps the instantaneous rate generated by the sender need to enforce N different window sizes and monitor the or received by the receiver below a specific level. The fun- amount of outstanding Transport Protocol Data Units (TPDUs) damental concept of enforcement of rate as the regulation to each of the N receivers. Furthermore, there is a real risk parameter is identical for both cases of unicast and multicast of acknowledgment implosion when using TCP for multicast communications. The regulation algorithms can however be communication where a few number of receivers experiencing high packet loss may trigger repeated retransmissions and This difference is especially important for the model-based slow down the entire multicast session. The sender is then case, where the feedback represents some measurement result forced to process a large number of acknowledgments from for some parameter that is used in model calculations. Ap- several group members only with the consequence that the propriate metrics for the evaluation of multicast traffic must sender may become a bottleneck for the whole multicast therefore be defined [9] as well as other parameters, like the group. This also has negative impacts on scalability, and is definition of fairness for rate-based regulation [41]. On the known as the ”crying baby problem” [22]. Another important other hand, in the case of increase/decrease algorithm, the problem is related to the need for good dimensioning of the feedback simply acknowledges whether there is congestion in group resources such as to reduce or eliminate the risk for Rate-based congestion control can be partitioned into sev- The mathematical foundation behind FEC is linear algebra eral classes, depending upon the place where the control over finite fields [11], [16], [19]. The n original segments are mechanism is implemented. These are [12]: viewed as the coefficients of a polynomial function of degree(n Source-based congestion control, where the source ad- − 1). Redundant segments can be generated by evaluating justs the transmission rate based on the information the polynomial function at m different points. Any n out of the received from the multicast receivers and/or based on m values fully specifies the polynomial, effectively regenerat- ing its coefficients. Two popular codes are the Reed-Solomoncode and the Tornado code [16]. Furthermore, FEC-based Receiver-based congestion control, which is generallyused in combination with layer-based multicast commu- packet recovery in the context of multicast communication can be done by inserting parity packets within a stream or acrossa combination of streams [26].
Hybrid congestion control, where both the source andthe receivers are participating in the congestion control E. Reliable AL multicast communication mechanism by reducing the rate (at the source) and thelayer subscription level (at the receivers), based upon the Overlay networks are opening for new facilities in multicast communication, the price however can be in terms of increasedlatency and also the risk for lower efficiency in resource TCP friendliness is achieved at rate-based mechanisms by utilization. Another important issue is regarding the reliable forcing the transmission rate to match a throughput that is multicast communication. Usually, this can be achieved by ”TCP compatible”, i.e., as given by the formula derived in applying TCP on the edges of a connection. Although this is [35]. A ”TCP compatible” flow is defined as a flow that a feasible solution, the price however can be high in form of, is responsive to congestion notification. Furthermore, this flow does not use, in steady state, more bandwidth than a AL multicast communication opens for more possibilities conformant TCP flow running under comparable conditions to do congestion and error control in a multicast group.
with reference to, e.g., loss rate, RTT, packet size.
There are two classes of control mechanisms in AL multicast Because of the acknowledgment implosion problem asso- communication, which are acting on end-to-end paths and hop- ciated with the window-based regulation, most of implemen- tations for reliable multicast communication use a rate-based End-to-end mechanisms means that congestion and error regulation mechanism to control and regulate the traffic [12], control are done on a end-to-end basis, irrespective of the number of hops. Congestion and error control mechanisms as D. Error control in multicast environments those described above can be used in this case for reliablemulticast communication.
Traditionally, error control mechanisms may use several On the other hand, hop-by-hop mechanisms means that techniques, and the most popular approaches are [12], [24]: congestion and error controls are done on a hop-by-hop basis Automatic Repeat ReQuest (ARQ) schemes, which use in an AL multicast group. An end-to-end path may consist acknowledgments, timers and retransmissions.
of several hops, each of which may include multiple unicast Forward Error Correction (FEC) algorithms, which en- able packet loss recovery at the destination provided that Hop-by-hop reliable AL multicast communication has been a specific number of packets are received non-corrupted.
shown to considerably reduce the average latency and jitter of Error Resilient Source Coding (ERSC), which is used to reliable communication [4]. This approach has also the advan- conceal possible errors at the receiver.
tage that it localizes congestion and error control mechanisms These error control mechanisms are suitable for specific to a specific subset of nodes and links in the overlay network.
applications and they can be used in connection with TCP or By this, loss recovery is localized, thus reducing the overall UDP. Delay-insensitive multicast applications (e.g., multicast link stress for packet retransmissions. Another advantage is bulk data transfer) have different time delivery requirements related to TCP friendliness, which can easily be implemented when compared to delay-sensitive multicast applications (e.g., in this case. The overhead induced by the hop-by-hop approach video distribution). For instance, in the case of multimedia have been shown to be insignificant [4].
distribution, reliable multicast communication means that the A possible drawback could however be the difficulties in delivery must be done reliably but also timely. FEC-based error applying the hop-by-hop scheme to global Internet due to recovery is therefore preferable for this kind of application.
scalability and interoperability issues.
FEC erasure correction restores corrupted packets by using other redundant packets [16]. There is also another form of FEC, so-called corruption correction, which corrects a cor- The paper has presented a survey on current solutions for rupted packet by using redundant information encoded within reliable multicast communication with emphasis on applica- the packet. Only erasure correction is relevant to transport tion layer multicast. These topics are subject for research protocols, because unrecoverable corruption is transformed within the research project ”Routing in Overlay Networks into erasure by the link or network layer.
(ROVER)”, granted in 2006 by EuroNGI NoE. The main focus of our research is on QoS-aware overlay routing for [22] Holbrook H., Singhal S. and Cheriton D.R., multicast environments, as a way to provide soft QoS pro- Reliable Multicast for Distributed Interactive Simulation, visioning for specific applications while retaining the best- [23] Internet Group Management Protocol, Version 2, RFC 2236, IETF, effort Internet model. Main research questions are on overlay http://www.ietf.org/rfc/rfc2236.txt?number=2236 multicast communication, traffic measurements and modeling, [24] IETF Working Group Reliable Multicast Transport (rmt), http://www.ietf.org/html.charters/rmt-charter.html QoS provisioning with multicast facilities and reliable multi- A Survey of Issues and Reliability and Congestion Control Techniques in Application Level Multicast, Planned future work is to develop a dedicated middleware Computer Science Dept., North Carolina State University, NC27606, USA [26] Kontothanassis L., Sitaraman R., Wein J., Hong D., Kleinberg R., environment, which will be used to develop new protocols Mancuso B., Shaw D. and Stodolsky D., A Transport Layer for Live for multimedia distribution over IP and to offer soft QoS Streaming in a Content Delivery Network, Proceedings of the IEEE, Vol.
guarantees for specific applications in a multicast environment.
[27] Lao L., Cui J-H., Gerla M. and Maggiorini D., A Comparative Study We are also planning to develop analytical and simulation of Multicast Protocols: Top, Bottom, or In the Middle?, IEEE Global Internet Symposium (GI 2005), Miami, FL, USA, March 2005 [28] Levine B.N. and Garcia-Luna-Aceves J.J., A Comparison of Reliable Multicast Protocols, Multimedia Subsystems, 6(5), August 1998 Multirate Video Multicast over the Internet: An Overview, IEEE Network, Vol. 17, No. 1, January/February 2003 [2] Al Hamra A. and Felber P.A., Design Choices for Content Distribution Retransmission (LVMR): Evaluation of Hierarchical Rate Control, IEEE in P2P Networks, ACM SIGCOMM Computer Communication Review, INFOCOM’98, San Francisco, CA, USA, March/April 1998 [31] Luby M., Gemmell J., Vicisano L., Rizzo L. and Crowcroft J., Asyn- [3] Amir Y., Awerbuch B., Danilov C. and Stanton J., chronous Layered Coding (ACL) Protocol Instantiation, IETF RFC3450, Flow Control for Reliable Multicast and Unicast in Overlay Networks, http://www.ietf.org/rfc/rfc3450.txt, December 2002 IEEE/ACM Transactions on Networking, Vol. 13, No. 5, October 2005 [32] Matrawi A. and Lambadaris I., A Survey of Congestion Control Schemes [4] Amir Y. and Danilov C., Reliable Communication in Overlay Networks, for Multicast Video Applications, IEEE Communications Surveys and IEEE International Conference on Dependable Systems and Networks Tutorials, fourth quarter 2003, Vol. 5, No. 2, 2003 (DSN03), San Francisco, CA, USA, June 2003 [33] Moen D.M., Pullen J.M. and Zhao F., Implementation of Host-Based [5] Biersack E.W., Where is Multicast Today?, ACM SIGCOMM Computer Overlay Multicast to Support of Web Based Services for RT-DVS, 8th Communication Review, Vol. 35, No. 5, October 2005 IEEE International Symposium on Distributed Simulation and Real-Time [6] Biersack E.W., Rodriguez P. and Felber P.A., Performance Analysis of Applications (DS-RT’04), Budapest, Hungary, 2004 Peer-to-Peer Networks for File Distribution, 5th International Workshop [34] Neumann C., Rocca V. and Walsh R., Large Scale Content Distribution on Quality of future Internet Services (QofIS’04), Barcelona, Spain, Protocols, ACM SIGCOMM Computer Communication Review, Vol. 35, [7] Braudes R. and Zabela S., Requirements for Multicast Protocols, IETF [35] Padhye J., Firoiu V., Towsley D.F. and Kurose J.F., Modeling TCP Reno RFC1458, http://www.ietf.org/rfc/rfc1458.txt, May 1993 Performance: A Simple Model and its Empirical Validation, IEEE/ACM [8] Castro M., Druschel P., Kermarrec A.-M. and Rowstron A.I.T., Scribe: A Transactions on Networking, Vol. 8, No. 2, April 2000 Large-Scale and Descentralized Application-Level Multicast Infrastruc- [36] Pallis G. and Vakali A., Insight and Perspectives for Content Delivery ture, IEEE Journal of Selected Areas in Communications, Vol. 20, No.
Networks, Communications of the ACM, Vol. 49, No. 1, January 2006 [37] Popescu Adrian, Erman D., Ilie D., Constantinescu D. and Popescu [9] Chalmers R.C. and Almeroth K.C., Developing a Multicast Metric, IEEE Alexandru, Internet Content Distribution: Developments and Challenges, GLOBECOM 2000, San Francisco, CA, USA, November 2000 4th Swedish National Computer Networking Workshop SNCNW2006, [10] Chu Y., Ganjam A., Ng T.S.E., Rao S., Sripanidkulchai K., Zhan J. and Zhang H., Early Experience with An Internet Broadcast System Based [38] Ramalho M., Intra- and Inter-Domain Multicast Routing Protocols: A on Overlay Multicast, USENIX Annual Technical Conference, Boston, Survey and Taxonomy, IEEE Communications Surveys and Tutorials, first [11] Comer D.E., Internetworking with TCP/IP Principles, Protocols and [39] Ratnasamy S., Francis P., Handley M., Karp R. and Shenker S., A Architecture, Volume 1, Pearson Prentice Hall, 5th edition, 2006 Scalable Content-Addressable Network, [12] Constantinescu D., Erman D., Ilie D. and Popescu A., Congestion and Error Control in Overlay Networks, technical report, Blekinge Institute PGMCC: A TCP-Friendly Single-Rate Multicast, SIGCOMM 2000, Stockholm, Sweden, August 2000 [41] Rubenstein D., Kurose J. and Towsley D., The Impact of Multicast NOSSDAV’03, Monterey, CA, USA, June 2003 Layering on Network Fairness, ACM SIGCOMM 1999, Cambridge, MA, Host Extensions for IP Multicasting, http://www.ietf.org/rfc/rfc1112.txt, August 1989 [42] Quinn B. and Almeroth K., IP Multicast Applications: Challenges and [15] Dragos I., Gnutella Network Traffic, Measurements and Characteristics, Solutions, IETF RFC3170, http://www.ietf.org/rfc/rfc3170.txt, Sept. 2001 Licentiate thesis, Blekinge Institute of Technology, ISBN 91-7295-084-6, [43] Saroiu S., Gummadi P.K. and Gribble S.D., Measuring and Analyzing the Characteristics of Napster and Gnutella Hosts, Multimedia Systems, [16] Dixit S. and Wu T., Content Networking in the Mobile Internet, John Wiley and Sons, Inc., ISBN 0-471-46618-2, 2004 [44] Shi Y.S., Design of Overlay Networks for Internet Multicast, [17] El-Sayed A., Roca V. and Mathy L., A Survey of Proposals for an thesis, Sever Institute of Technology, Washington University, Saint Louis, Alternative Group Communication Service, IEEE Network, Vol. 17, No.
[45] Vicisano L., Rizzo L. and Crowcroft J., TCP-Like Congestion Control [18] Eriksson H., MBONE: The Multicast Backbone, Communications of for Layered Multicast Data Transfer, IEEE INFOCOM’98, San Francisco, [19] Forouzan B.A., TCP/IP Protocol Suite, 3rd edition, McGraw-Hill, 2004 [46] Wang J., Yurcik W., Yang Y. and Hester J., Multi-Ring Techniques [20] Ganjam A. and Zhang H., Internet Multicast Video Delivery, Proceed- for Scalable Battlespace Group Communications, IEEE Communications ings of the IEEE, Vol. 93, No. 1, January 2005 [21] Gummadi K., Gummadi R., Gribble S., Ratnasamy S., Shenker S. and The Impact of DHT Routing Geometry on Resilience and Proximity, ACM SIGCOMM’03, Karlsruhe, Germany, August 2003

Source: http://dragos.ownit.nu/pubs/popescu2007-survey_reliable_multicast.pdf