Leading group [ YU HU & Mindy ]

Topologically-Aware Overlay Construction and Server Selection

About this paper, we discussed the main idea of binning scheme and its
useness
in applications: Overlay Network Construction and Server Selection.
We discussed:
1.The details of CAN Contruction and its uneven distribution. 
2.The possible relation and impact between distribution of IP level and
overlay network. 
3.How to decide the number of landmarks and these landmark machines
position.
4.When internet changed much dynamically , how can we do to keep update
information and suitable landmarks for internet .
And then, we want to discuss about: 
1. The measurement is on latency, is there some scheme on other
measurement?
2. Is there some mechanism that decide issues on landmark machines based
on itsmeasurement of the network ?


A Scalable Content-Addressable Network

The paper has a detailed research in Content_Addressable network and
demonstrates its scalability, robustness and low-latency properties
through simulation. They addressed two key problems in the design of
CANs: scalable routing and indexing. The authors argued that large-scale
distributed systems could benefit from hash table functionality like
software systems. They also explain one possible application of CAN,
i.e. "grass-roots" content distribution system. They first constructed
their design and then did certain improvements on that. This evolusional
method really works well and is practical and useful.

Discusses:
We discussed the benefits and drawbacks of CAN. Some comparisons between
this paper and Chord paper were argued. As to the two papers are from the
same group, so their research was discussed more in detail.

The strongest and/or most interesting ideas in the paper:
a. Large-scale distributed systems could benefit from hash table
functionality like software systems.
b. Use multiple hash functions to improve data availability

The most striking weaknesses in the paper:
a. Some evasions because of the limitations in their research, e.g. they
denied to decide on tradeoffs between improved routing performance and
system robustness on the one hand and increased per-node state and system
complexity on the other because they had no enough understanding of the
requirement.
 
b.They made some quantity analysis to make their argument sound.  But they
can be more convinced if they built their simulation on different
topologies and did some comparative work.

Questions:
a. So far, are there some research results about the tradeoffs between
improved routing performance and system robustness on the one hand and
increased per-node state and system complexity on the other hand?

b. Which is more favorable of this paper and the Chord paper?

c. How's about the conclusion if the cost is considered?