[CSC2231] Paper Review: Glacier: Highly Durable, Decentralized Sorage Despite Massive Correlated Failures

Glacier is a P2P storage system with massive redundancy to maintain data
durability. Glacier maintains a small number of full replicas and also
distributes fragments of data to various hosts for long-term durability.
Both durability and storage overhead can be expressed as a function of
node failure rate, number of replicas and number of fragments.

Maybe because of the various parameters involved, the authors do not
show the relationship or the trade-off between data durability and
storage overhead directly. I feel this is important because the authors
emphasize "massive redundancy" is the key idea in this work.

I am not convinced by the idea of having both replicas and fragments in
a P2P system. Under this environment, I think all hosts that request
data will go for the replicas because there are less hassles to do so.
The fragments will eventually function as the source of data to repair
the replicas when a replica node fails.
Received on Thu Nov 24 2005 - 11:01:51 EST