18
Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA), Dario Mag giorini (Uni. of Milano)

Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Embed Size (px)

Citation preview

Page 1: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005

A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle?

Li Lao (UCLA), Jun-Hong Cui (UCONN)

Mario Gerla (UCLA), Dario Maggiorini

(Uni. of Milano)

Page 2: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 2

Why Another Study?

Multicast solutions have been evolving from “bottom” to “top” IP multicast, application layer multicast, overlay multicast Incredible amount of research …

Little research has been done to systematically compare the performance of different layer protocols How much worse are upper layer solutions? Are they long-term substitute to IP multicast or temporary

solutions? How will overlay design impact overlay multicast performance? Which architecture should we choose in which scenario?

Page 3: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 3

Outline

Multicast Overview Experimental Methodology Simulation Studies Conclusions

Page 4: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 4

Multicast Overview

IP Multicast

Application Layer Multicast Overlay Multicast

Page 5: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 5

IP Multicast

Relies on network routers Pros

Bandwidth efficiency Cons

Lack of scalable inter-domain multicast routing protocols Require global deployment of multicast-capable routers Lack of practical pricing models

Examples: DVMRP/PIM-DM, CBT, PIM-SM, MOSPF, PIM-SSM, …

Page 6: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 6

Application Layer Multicast (ALM)

Relies on end systems only Pros

Ease of deployment Cons

Lower bandwidth efficiency and higher end-to-end delay Heavy control overhead Challenges for large groups

Examples: Yoid, ESM, ALMI, NICE, TAG, HyperCast, …

Page 7: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 7

Overlay Multicast (OM)

Relies on intermediate proxies to form a “backbone” overlay Pros

Implicitly gains knowledge about the network topology More efficient group management Reduced control overhead Support multiple groups/applications simultaneously

Cons Deployment and maintenance cost of overlay proxies Requires careful design of the overlay network

Examples: Scattercast, Overcast, RMX, AMCast, OMNI, …

Page 8: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 8

A Qualitative Comparison

Metrics IP ALM OM

Ease of Deployment Low High Medium

Multicast Efficiency High Low Medium

Control Overhead Low High Medium

Page 9: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 9

Experimental Methodology (I)

Topology graphs Router-level: Rocketfuel (University of Washington) AS-level: Route Views (University of Oregon)

Group membership generation Uniform distribution

Multicast Protocols IP multicast: PIM-SSM Application layer multicast: Narada and NICE Overlay multicast: POM (Pure Overlay Multicast)

End users connect to proxies via unicast

Page 10: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 10

Experimental Methodology (II) Overlay design

Overlay proxies: nodes with the highest degree Overlay links: adjacent connection

Performance Metrics Multicast tree quality

Tree cost: number of physical links in multicast tree End-to-end delay: # of hops between source & receivers

Control overhead Tree setup/tear-down, tree refresh, overlay link measurement For a single group and multiple groups

Reliability, stability, security, etc.

Page 11: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 11

Multicast Tree Cost

OM has lower cost than ALM Among ALM, NARADA outperforms NICE for small groups, but not for

larger groups Tree cost of POM increases faster than IP and ALM

Can use IP or ALM instead of unicast between proxies and end users

Page 12: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 12

End-to-End Delay

OM has slightly higher latency than IP multicast Among ALM, the delay of NARADA remains fairly constant, and the delay of

NICE increases very rapidly Trade-off between multicast tree cost and end-to-end delay

Page 13: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 13

Control Overhead (I)

IP multicast has lowest overhead overall ALM has less overhead than OM for small groups, but its overhead exceeds

OM when group size increases beyond a point OM curve has a smaller slope than ALM curves

Backbone overlay maintenance overhead is independent of group size

Page 14: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 14

Control Overhead (II)

Backbone overlay maintenance overhead is independent of the number of groups

Control overhead of ALM is proportional to the number of groups

Page 15: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 15

Impact of Overlay Parameters

A larger number of proxies help to reduce multicast tree cost

Page 16: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 16

Impact of Overlay Parameters

Backbone overlay maintenance overhead increases with the number of proxies

Page 17: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 17

Conclusions

Application layer multicast A suitable solution for immediate deployment Good for small groups

Overlay multicast Could achieve performance comparable to IP multicast A good choice for large numbers of groups Could serve as a long-term solution

Future work Reliability, stability, security… Different group membership models Overlay network design

Page 18: Global Internet 2005 A Comparative Study of Multicast Protocols: Top, Bottom, or In the Middle? Li Lao (UCLA), Jun-Hong Cui (UCONN) Mario Gerla (UCLA),

Global Internet 2005 18

No more questions, please!