Review - High Availability, Scalable Storage, Dynamic Peer Networks: Pick Two

This paper brings to light the fact that maintaining redundancy in a high
churn peer-to-peer system is costly. As such, maintaining high availability
in a dynamic system, providing storage guarantees and expecting good
scalability cannot reasonably be achieved at the same time. The limiting
factor in this result is upstream peer bandwidth.

The authors show that, because transient node availability is low, an
increase in redundancy is needed to maintaining content availability. Since
it is well understood that peer-to-peer nodes join and leave frequently,
this is an intuitive result. Moreover, because nodes are only available for
a fraction of the time, the bandwidth they contribute is reduced. These
results illustrate relationship between redundancy and availability.

While this paper demonstrates several important aspects of peer-to-peer
churn, Blake and Rodridges make several unrealistic assumption throughout
the paper. For example, they discuss nodes that are required to initially
download all the content that they will serve. These assumptions are
justified by claiming that they are conservative, but it seems that some may
result in a bandwidth increase.

While I believe there are several inaccuracies in this report, the model
presented allows for a simplified analysis. Further, it convincingly
demonstrates the relationship between peer availability, redundancy and
bandwidth.
Received on Mon Nov 14 2005 - 09:53:22 EST