QoSMIC: a Quality Of service Sensitive Multicast Internet protoCol

1
QoSMIC a Quality Of service Sensitive Multicast
Internet protoCol
Michalis Faloutsos Anindo Banerjea Rajesh Pankaj
2
Motivation

• Modern applications
• Tele-conferencing
• Tele-education

• Current Internet
• Packet switching
• Best-effort approach

3
Desirable Protocol Properties

• -- Alternate paths
• -- Dynamic routing information
• Efficiency
• Application sensitive
• Scalability
• Robustness
• Loop-freedom

4
Current Protocols BGMP

• Border Gateway Multicast Protocol
• No QoS support
• Reverse Shortest Paths Routing

5
Current Protocol YAM

• Yet Another Multicast protocol
• Multiple paths
• Use of static information
• Not scalable
• -- flood of control messages

6
This is just not effectiveWe need to get some
chains

• The Far Side -- G. Larson

7
QoSMIC

• QoS Multicast Internet protoCol
• Supports QoS
• Uses dynamic information
• Scalable

8
QoSMIC The Overview

• Manager Router
• 1. Search for Candidates
• 2. Bid Candidates Bid
• 3. Select New chooses path
• -- using dynamic QoS info.

9
QoSMIC Search and Bid

• Local Search (costly)
• Multicast Tree Search
• Bid messages collect
• Connection Establishment

10
QoSMIC Local Search

• New router floods in neighborhood
• -- reverse path multicasting
• -- scope controlled by TTL
• Candidate routers reply
• -- dynamic QoS metrics
• Bid messages collect

11
QoSMIC Multicast Tree Search

• New router unicasts to Manager
• Candidate Selection
• -- centralized selection
• -- distributed selection
• Bid messages collect

12
QoSMIC Proof of Correctness

• Loop-free distribution
• -- Lemma 1 A tentative path is loop-free
  independently of the underlying unicast protocol
• Proof router(s) with tentative state
  discard the Bid for the
• same.
• -- Theorem 1 The routing table information on all
  the routers for a group (Shared Tree) or for a
  source of a group (Source-Based Tree) constitutes
  a forest at every instant.
• Proof Take-Over procedure

13
Control Overhead Comparison

• Complexity of Selecting Candidates

• Complexity of Joining

14
Simulation Goals

• Compare routing efficiency with BGMP
• -- Tree cost Sum of link-weights
• Compare message complexity with YAM
• -- Number of control messages of search
• phase

15
The Simulated Algorithms

• Offline greedy for reference
• SP Shortest Paths
• BGMP Reverse Shortest Paths
• QoSMIC
• Def. Group density group size/network size
• Def. Maximum asymmetry as the maximum
• ratio of the opposite edges between a pair
  
• of routers

16
Tree Cost Vs Group Size
Asym 10

• QoSMIC up to 40 better than BGMP

17
Tree Cost Vs Asymmetry
Group density 18

• QoSMIC up to 60 better than BGMP

18
Joining Distance Distribution
Group Density 1-15

• Even small Local Search is good (2 h - 55 joins)
• Large Local Search (8 h) to cover for ALL cases

19
Search Message Complexity

• YAM has to have large Local Search
• QoSMIC can keep LS small, thus scalable

20
Summary QoSMIC Profile

• Pros
• -- Support QoS
• -- Scalable
• -- Create efficient trees
• -- Adaptive - Flexible
• Cons
• -- does not address end-to-end QoS