6
Scalable Reliable Multicast Architecture Wenjun Zeng Computer Science Department University of Missouri- Columbia

Scalable Reliable Multicast Architecture

  • Upload
    sloan

  • View
    38

  • Download
    1

Embed Size (px)

DESCRIPTION

Scalable Reliable Multicast Architecture. Wenjun Zeng Computer Science Department University of Missouri-Columbia. Hybrid IP Multicast and ALM. IP Multicast (NM) Pros: bandwidth efficient, low delay, more stable Cons: deployment issue, reliability/scalability issue ALM: - PowerPoint PPT Presentation

Citation preview

Page 1: Scalable Reliable Multicast Architecture

Scalable Reliable Multicast Architecture

Wenjun ZengComputer Science Department

University of Missouri-Columbia

Page 2: Scalable Reliable Multicast Architecture

2

Hybrid IP Multicast and ALM IP Multicast (NM)

Pros: bandwidth efficient, low delay, more stable Cons: deployment issue, reliability/scalability issue

ALM: Pros: super scalability Cons: less bandwidth efficient, potential longer delay, unstab

le (high node dynamic) Hybrid NM and ALM

Bridge between NM and ALM, e.g., hybrid tree construction Adaptation How about other issues (reliability, congestion control, etc)?

Page 3: Scalable Reliable Multicast Architecture

3

Push or Pull? IP Multicast:

Push ALM:

Tree-based approaches: push (e.g., End System Multicast) More efficient but has instability issue (and bandwidth underutilization)

Data-driven randomized approaches: pull (e.g., CoolStreaming)

No tree, simple, better suited for addressing reliability but incur latency-overhead trade-off

Hybrid overlay: push and pull E.g. tree-bone based

Hybrid NM and ALM? IP multicast tree bone (push)? Address instability! Overlay (tree or mesh) for other issues, e.g. scalable

reliability?

Page 4: Scalable Reliable Multicast Architecture

4

Example: Reliability: IP Multicast Retransmission

Potential problems: Reliability?

It is likely that retransmitted packets will go through same (probably congested) routes as the original packets, likely to experience congestion and loss too. (bottleneck problem)

Retransmitted packets may deteriorate the congestion. Scalability?

In an IP multicast case, the retransmitted packets will be sent to all participants, wasting bandwidth.

Although separate unicast session can be used, it increases the load of the original sender.

Page 5: Scalable Reliable Multicast Architecture

5

Path Diversity Overlay Retransmission

To exploit path diversity, each client identifies a couple of “good” Retransmission Nodes in overlay.

Overlay rtx network (pull)

IP Multicast network (push)

Sender

Receiver

Rtx node

Page 6: Scalable Reliable Multicast Architecture

6

Observations

A pull-based (data-driven) overlay architecture seems to have more flexibility in exploiting path diversity to reduce the retransmission delay and to avoid severely congested (bottleneck) path. to provide load balancing

It may be built on top of the traditional IP multicast architecture, ALM architecture or some hybrid