Implementation of Host-based Overlay Multicast in Support of Web - - PowerPoint PPT Presentation

Implementation of Host-based Overlay Multicast in Support of Web Based Services for RT-DVS

Dennis Moen, Mark Pullen & Fei Zhao

George Mason University {dmoen,mpullen,fzhao}@gmu.edu

Network Service Requirements for Real Time Distributed Virtual Simulation

Network Quality of Service (QoS)

end-to-end capacity, latency, jitter, and packet loss in a statistical sense

Multicast

many-to-many group communication

Reliable Multicast Transport

high confidence of delivery

End-to-end network status and performance monitoring

need to know what the network is doing for you

Multi-sensor systems

must manage streaming data with low latency

Internet Multicast Services Today

• IP multicast over the Internet not widely deployed
• IETF initial focus is on one-to-many multicast
• Commercial viability lacking for IP multicast in the

Internet

• Result: interest in multicast based on end

systems not network

– End-to-end argument: push complexity up the stack – Example: TCP is complex, IP is simple

INTERNET INTERNET

A C B G E F D

IP Multicast tree:

A C B G E F D H J

Overlay Multicast Tree

Router A XOM1 Router B XOM4 XOM2 XOM3 Router C XOM5 Router D XOM6 XOM7

XOM Overlay

Generic Class Definition Interface (SRMP Example) Packet Send/Receive Distribute Messages Listen to Ports Class QoS/ Queueing Routing Table Group Management Registry Join/leave Security Address Capacity/latency Node Demand Path Optimization Path Management Routing UDP IP

XOM Layers

Application B sending implies routing to group G3 = {G1? G2} Internet XOM1 A B C D XOM2 A XOM3 B C D

G1 = {A, B}

G2 = { B, C, D} B

G1 = {A, B}

G2 = { B, C, D}

XOM Group Membership

Group Aggregation Overlay

(Optimum Path Overlay)

Multicast Groups Group Members g0 XOM1,2,3,4 g1 XOM1,2,3,4 g2 XOM1,2,3 g3 XOM1,2 Aggregate Trees Tree Tree Links (arcs) T0 1-4, 4-2, 4-3

XOM2 XOM4 XOM3 XOM1

Internet

(g0, g1, g2, g3) (g0, g1) (g0, g1, g2) (g0, g1, g2, g3) T0

Groups g0, g1, g2, g3 share one aggregate tree T0. T0 is a perfect match for g0 and g1, but is a leaky match for g2 and g3. Trades off path utilization inefficiency for lower path management

• verhead.

XOM Simulation Application End-to-End Latency Path Constraint Demand Constraint Minimum tree Optimum Path Traffic Load XOM Internet Access capacity (Rate Control)

Overlay Routing Constraints

UDP IP Packet Sender Host Channel Abstraction for Multicast Channel (S,G) Routing Table for Channel (S,G) Traffic Generator SRMP Packet Receiver

XOM Functional Model

Prototype Test Scenario

XOM Prototype

Statistics Internet Registries Other XOM Sites MulticastRouter* (Java or C++) Routing IPmc Host IPmc Host

WAN stats LAN stats data incoming outgoing routing info routing table multicast to/from WAN *All modules except Router are Java

Test 2. XOM n-degree of 2

XOM

2

XOM

3

XOM

1

XOM

4

XOM

2

XOM

4

XOM

3

XOM

1

Test 1. XOM n-degree of 3

XOM Lab Test Scenarios

0.0 10. 20. 30. 40. 50. 60. 70. 80. 90. 500 1000 1500 2000 2500 Messages/sec Delay (msec) 2-degree 3-degree .

Message Delay

0.00% 1.00% 2.00% 3.00% 4.00% 5.00% 6.00% 500 1000 1500 2000 Messages/sec Loss Ratio (%) 2-degree 3-degree

Message Loss Ratio

Conclusions and Future Work

Initial results indicate overlay networking is a promising strategy for providing many-to-many multicast in the open network environment of DS-RT. We are working on an architecture specification based

• n the properties of distributed simulation traffic plus

recent networking research. NPS is working on a Web-service-based registry and routing information system.