Paper Review: Enabling Conferencing Applications on the Internet using an Overlay Multicast Architecture
Reviewer: Kevin Hofstra
1. How can we efficiently enable conferencing on a small yet reliable scale over a largely distributed network such as the internet? Should this be optimized for throughput or for latency? What are the limitations for number of members for the multicast group or other possible limitations?
2. An alternative method for group communications over the internet. Investigation into how application requirements will affect End System Multicast Design. A test of the architecture through the use of 20 nodes distributed sparsely and geographically diversely over the internet.
A. New method involves putting all multicasting related functionality such as group management and packet replication at the end systems instead of using the IP layer.
B. They use a redesigned Narada protocol. Data delivery is performed by organizing into overlay spanning trees. Each link corresponds to a unicast overlay of 2 systems in the underlying internet. This has many advantages and disadvantages:
i. Routers do not need to keep any state (increased scalability).
ii. No network infrastructure changes must be made to implement (ease of implementation).
iii. High level features can be implemented over the necessary links such as FEC and congestion control (increased functionality).
iv. Must effectively self organize into an overlay tree network that is efficient. (difficult and costly)
C. A method of optimizing for either throughput or latency makes the architecture more robust depending on your needs. In their testing the optimal technique for creating efficient overlays was bandwidth-Latency. It has performance comparable to Sequential Unicast and sometimes even performed better.
4. Rating- 4
The architecture they explore seems to very effective for its purposes. However it is fairly limited in its scope as far as which applications that it may be effective for.
The evidence presented in their trials seems very convincing. They often tell admit and describe the shortcomings in the design and do not seem to be hiding anything. However they say that it may be used for 10s to 100s of group members yet only test on 20, which makes me skeptical that the group members can make an effective overlay tree with larger numbers of members.
5. Systems analysts should recognize that this architecture does not try to make any network infrastructure changes or try to extend beyond its limited scope. The architecture admits its flaw of need its members to self create an effective overlay tree and therefore limits its scope to 10s or 100s of members. Instead of trying to make it feasible for larger groups they recognized the niche that it fills and keeps it simple and effective without trying to overstep its intended area. I think the reason this paper is effective is because the architecture fills its intended role, and its intended role only, but quite well.