[CSC2231] Paper Review: Total Recall: System Support for Automated Availability Management

This paper looks at the eagerness to replicate data in a P2P storage
system. This is a trade-off between availability and bandwidth because
eager replication is necessary to maintain availability but consumes
bandwidth at the same time. When P2P hosts are highly dynamic, it is
better off to retreat from urgent repairs to prevent the repair copy of
data replicated to the hand of transient hosts.

I am not sure if the paper addresses the relation between the number of
required replicas (c) and the eagerness to repair. Intuitively, the less
copies there are in the network (as suggested by erasure-coded
replication scheme in equation 5), the higher the eagerness to repair.
The evaluation of this work relies on the erasure-coded scheme to
compute c. However, if the P2P storage system use another way to
calculate c, such as using equation 2, will the results of this work change?

As mentioned in class, the availability of a distributed system depends
on the relative availability between the hosts. If host A, which serves
data, is up while host B, which demands data, is down, the availability
is zero no matter how high of availability of host A maintains.
Therefore, I'd suggest the P2P replication research looks further into
ways of distributing replicas of data to hosts that are available at
different time of the day, and using this idea to migrate data between
hosts during repairs.
Received on Thu Nov 24 2005 - 10:06:15 EST