Optimizing Multicast Throughput in IP Network
1 like1,278 views
The document summarizes research on optimizing multicast throughput in IP networks. It proposes tree packing algorithms that can achieve throughputs close to the theoretical maximum allowed by network coding. Specifically, it introduces a Cooperative Shortest Path Tree Packing Algorithm (CSPT) for distributed tree packing that works well by avoiding edge congestion. The CSPT approach is implemented on standard routing protocols like OSPF to evaluate its performance through simulations.
![34
Some Simulation Results
S Si l ti R lt
• Random graph with 50 nodes uniform link capacity
nodes, uniform link capacity
ranged [1,10] G is considered and network degree from 2
to 10.
• Up to 5 source nodes and total number of 30 terminals
are considered.
Optimizing Multicast Throughput in IP Networks](https://image.slidesharecdn.com/thesisipmulticast-120615162916-phpapp01/85/Optimizing-Multicast-Throughput-in-IP-Network-34-320.jpg)
Optimizing Multicast Throughput in IP Network
- 1. 1 Optimizing Multicast p g Throughput in IP Networks M. Reza Rahimi, M. Reza Rahimi Software Systems Engineering, University of Regina, Canada, September 2009. September 2009.
- 2. 2 Outline O tli • Introduction • Thesis Goal • Problems Formulation • Proposed Tree Packing Algorithms • Implementation On Standard Protocols • Some Simulation Results • Conclusion and Future Directions Optimizing Multicast Throughput in IP Networks
- 3. 3 Introduction I t d ti • Data transfer over network could be categorized into g three main groups: • Unicast: • Transfer of data from one source to one destination (Path Packing). Source: Wikipedia • Broadcast: • Transfer of data from one source to all of the entire nodes in the network (Spanning Tree Packing ). Optimizing Multicast Throughput in IP Networks Source: Wikipedia
- 4. 4 • Multicast: • Transfer of data from one source to group of destinations but not all and not one of the nodes (Steiner Trees). Source: Wikipedia • Different Optimization Metrics Optimization Metrics can be considered depending on the application: • Maximum amount of information into terminals (Internet). M i f i f i i i l (I ) • Energy (Wireless Sensor Networks). • Delay (Internet Telephone). • Fault Tolerance (specially in wireless networks). ( p y ) Optimizing Multicast Throughput in IP Networks
- 5. 5 • In this thesis we are concerned with the Throughput Throughput (optimum data transfer ) (optimum data transfer ) and Quality in Multimedia Quality Multimedia Applications. • Let’s consider the main three mentioned problems in much more details. • This will lead us to the following questions: • Question: • What is the Maximum Amount of Information that can What is the Maximum Amount of Information that can be transferred in each scenario? Is the solution traceable in reasonable time ? Optimizing Multicast Throughput in IP Networks
- 6. 6 • Unicast: The answer is yes according to the Maximum‐ Maximum‐ Maximum Flow Min‐Cut Theorem: Flow Min‐Cut Theorem We can find the maximum possible information transfer rate with routing (only forwarding information packets) in polynomial time in Unicast Scenario. • Broadcast: According to the Edmond’s Theorem we can Edmond’s Theorem find the optimum solution in polynomial time. • S h So in this scenario, with routing and duplication h h routing and duplication the d d l optimum value can be reached in polynomial time. • Multicast: Unfortunately the following result is valid : Multicast: Unfortunately , the following result is valid : Optimizing Multicast Throughput in IP Networks
- 7. 7 • Generally there is no Polynomial Time Generally there is no Algorithms to find the optimal Duplicate and to find the optimal Duplicate Forward strategy in Multicasting (P≠NP). strategy in Multicasting (P≠NP). Optimizing Multicast Throughput in IP Networks
- 8. 8 Network Coding vs Routing N t k C di R ti Optimizing Multicast Throughput in IP Networks
- 9. 9 • This technique is called network coding network coding in literature. • Network coding contains routing as a special case The optimal network coding throughput is at least as large as the optimal routing throughput • Question: How can we find the best coding scheme and what is the maximum possible throughput? p g p • The first question is hard to answer in many different cases. But there is a useful technique called CFLow which CFLow results the throughput of network coding. lt th th h t f t k di • This technique is used to find the upper bound on g g p routing throughput Optimizing Multicast Throughput in IP Networks
- 10. 10 Thesis Goal Th i G l • We consider the generalized version of multicast when there are Several Sources Several Sources in the network. • We consider many different fairness criteria, for y , example optimizing the minimum throughput among sources and sources with unequal demands demands. • B th d t t Both data transfer application and multimedia f li ti d lti di application are considered. • Although network coding could increase the throughput of some network dramatically, it has been shown that for Undirected Networks (network with full Undirected Networks (network with full duplex communication links ) duplex communication links ) this gap is small. duplex communication links ) this gap is small Optimizing Multicast Throughput in IP Networks
- 11. 11 • We proposed tree multicast routing optimization algorithms based on standard IP routing protocols, that can achieve throughputs very close to the network coding upper bound (92% in average). coding upper bound (92% in a erage) • For multimedia applications rate‐distortion , y framework, a metric that is usually used for measuring g multimedia application performance, is used. • Through simulation we show that our algorithms achieve average multimedia multicast qualities that are only on average, 0.44 db worse than that achievable with network coding Optimizing Multicast Throughput in IP Networks
- 12. 12 Problems Formulation P bl F l ti • Let be a weighted undirected graph . weighted undirected graph • Now define to be a set of source nodes , where | | is the set cardinality operator. | | y p • Let be the set of sink nodes of source Si. • as the set of all Steiner trees rooted at Si that have as the terminal points . • A demand vector is said to be achievable if there exists a rate assignment function such that: Optimizing Multicast Throughput in IP Networks
- 13. 13 • The first constraint is the link capacity constraint link capacity constraint that the total flow on each link must not exceed the link th t t l fl h li k t t d th li k capacity, • The second one is the source throughput or demand throughput demand. • These constraints are the only constraints when tree packing is allowed. Optimizing Multicast Throughput in IP Networks
- 14. 14 • MaxMin Multicast MaxMin Multicast optimization problem: • When the goal is to maximize the minimum demand of the source nodes: • If network coding is allowed, using Cflow technique, the following program finds the upper bound on MaxMin optimization problem: Optimizing Multicast Throughput in IP Networks
- 16. 16 • Linear Multicast Optimization Problem: • Different rate priorities to different sources. • And its network coding version: Optimizing Multicast Throughput in IP Networks
- 18. 18 • Rate Distortion Optimization Problem: • This frame‐work is suitable for multimedia applications. pp • D is the rate‐distortion function. In this thesis, a power‐ is the rate distortion function. In this thesis, a power law function will be used. • The resulting optimization problem when network coding is allowed: Optimizing Multicast Throughput in IP Networks
- 20. 20 Proposed Tree Packing Algorithms P d T P ki Al ith • Three‐Classes of Cooperative Tree‐Packing Algorithms: Three‐Classes of Cooperative Tree‐ p g g • Non‐Cooperative Tree Packing Class: In this class each Non‐Cooperative Tree Packing Class source packs its own tree independently (Greedy), according to some strategy (e.g., minimum hop tree, minimum weight tree, maximum weight Steiner tree). i h S i ) Optimizing Multicast Throughput in IP Networks
- 22. 22 • Medium‐Cooperative Tree Packing Class: Medium‐ • Round‐robin family of tree packing algorithms. • In each round of the algorithm, each source selects one tree according to its own strategy. g gy • The rate assignment is postponed to the end of each round, i.e., when each source has nominated one tree in that round. one tree in that round • At the end of the each round, we know how many times each edge has been selected. Thus, the maximum rate that could be assigned to each th i t th t ld b i d t h edge is equal to the capacity of that edge, divided by the number of times it has been selected Optimizing Multicast Throughput in IP Networks
- 24. 24 • Highly‐Cooperative Tree Packing Class: Highly‐ • the Rate assignment is postponed until the very last round. • Sources pack one tree at each round, according to their own strategy. At the end of all the rounds, rates are assigned to the trees. g • This class of algorithms is our main proposed solution to the distributed tree packing problem. • W We propose the Cooperative Shortest Path Tree th C Cooperative Shortest Path Tree C ti Sh t t P th T Packing Algorithm (CSPT CSPT) for this purpose. Optimizing Multicast Throughput in IP Networks
- 27. 27 Why CSPT works well? Wh CSPT k ll? Optimizing Multicast Throughput in IP Networks
- 31. 31 Implementation On the Standard Protocols • Using OSPF as the basic protocol as the basic protocol. • OSPF routing protocol is a Link State protocol. • It is based on cost rather than hops (hops could be considered as the special case of cost). • All the OSPF routers has the Link State Database (LSDB) and executes Dijkstra's algorithm on their database to calculate a shortest path route to a given destination node from the current router. • The routers database information are periodically sent to the entire routers in the network. Optimizing Multicast Throughput in IP Networks
- 32. 32 • In OSPF, two important multicast addresses are used. • When an OSPF area is started, one router is elected the Designated Router (DR), and another as the Backup Designated Router (BDR) • The Designated Router tells all the other routers about changes in the network by sending out Link State Advertisements (LSA) on multicast address 224.0.0.5. • Every change in the network topology will be send by the OSPF routers on multicast address 224.0.0.6, the OSPF routers on multicast address 224 0 0 6 reserved for the DR and BDR. Optimizing Multicast Throughput in IP Networks
- 33. 33 • Using OSPF , 3 classes of tree packing could be implemented as follows: i l d f ll Optimizing Multicast Throughput in IP Networks
- 34. 34 Some Simulation Results S Si l ti R lt • Random graph with 50 nodes uniform link capacity nodes, uniform link capacity ranged [1,10] G is considered and network degree from 2 to 10. • Up to 5 source nodes and total number of 30 terminals are considered. Optimizing Multicast Throughput in IP Networks
- 35. 35 • Simulation Results of the MaxMin Multicast Optimization Problem : O i i i P bl • The throughput is defined as: • The following averaged results obtained: CSPT Cooperative Shortest Path Tree NCSTEIN NON-COOPEARIVE STEINER MCSTEIN MEDIUM – COOPERTIVE STEINER MCDIJK MEDIUM- COOPERTIVE DIJKSTRA MCBFS MEDIUM- COOPERTIVE BFS NCBFS NON-COOPEARIVE STEINER NCDIJK NON-COOPERATIVE DIJKSTRA Optimizing Multicast Throughput in IP Networks
- 36. 36 • Convergence Speed: • The relation of the average number of trees and the average percentage of the CSPT, NCSTEINE and MCSTEINE throughput to the network coding g p g throughput: Optimizing Multicast Throughput in IP Networks
- 38. 38 • Linear Multicast Optimization Problem: • It reaches on average to the 90% of network coding. • The following shows the examples with some priority vector. Optimizing Multicast Throughput in IP Networks
- 39. 39 • Simulation Results of the Rate Distortion Optimization Problem : Optimizing Multicast Throughput in IP Networks
- 41. 41 Conclusion and Future Directions C l i d F t Di ti • In this thesis the problem of IP multicast in multi In this thesis the problem of IP multicast in multi‐ source environments is discussed and formulated. • We have proposed a novel tree packing algorithm called p p p g g CSPT which has the throughput close to the network coding or in average about 92% of the network coding. • F lti di applications, CSPT has 0.44db diff For multimedia li ti CSPT h db difference with network coding performance • With packing 8 trees per source using CSPT, one could reach the throughput of 50% of network coding for data transfer applications and 3.27db different with network coding for multimedia applications. coding for multimedia applications Optimizing Multicast Throughput in IP Networks
- 42. 42 • There are some issued that should be considered: • Directect network should be studied. • Noisy network should be considered. • For multimedia application, correlated source should be considered. Optimizing Multicast Throughput in IP Networks