CSC2231 - Pick two review

High Availability, Scalable Storage, Dynamic Peer Networks: Pick Two
--------------------------------------------------------------------
C. Blake et al.

The paper tries to explain what the scalability problem in P2P
systems really is. By building a model and by looking at the behavior
of the Gnutella file-sharing network, the authors show that the real
issue in a redundant P2P architectures is bandwidth.

As paper weaknesses, I don't quite get why they assumed a constant
rate of joining and leaving. In P2P systems, isn't it the case of
highly dynamic membership?

Another thing is the statement about the six nines availability. I
feel that part is not useful and wrong mainly because is not impossible
to get that availability in a system that depends on other architectures
(DNS for example) that have 2 or 3 nines availability.

Also, the authors state that data redundancy is the key to any
data guarantees. However, in my opinion, content popularity should
be the main factor in the replication process. The paper doesn't
discusses that. For example a popular content is very probable
to be on a high number of peers so why worry about redundancy
in this case? In the same manner, the least popular objects do
not matter that much.
Received on Mon Nov 14 2005 - 10:19:42 EST