Managing the Data Replicas Efficiently in Time-Zoned Multilayer Chord

Replica is better than erasure coding when managing the redundancy of a storage system under high churn. It needs extra network communication to keep the target redundancy. By classifying the hosts into different categories of availability and placing the data replicas carefully may reduce the maintenance overhead. Considering a globe scale P2P system, the time-zoned multilayer Chord divides the hosts into 24 categories according to the time zone where they located. All the hosts are belongs to a Main Chord which consisted of all the hosts and one Local Chord which consisted of the hosts located in the same time zone. Because of the daily behavior of the people in same time zone are nearly similar, which means their computers are more likely to be available in the daytime and less likely to be available in the nighttime. So, for a certain time, there is nearly half of the hosts get higher availability than the other hosts. As the hosts in the same time zone are considered, their availability evaluated by the system is shifting periodically in a 24 hours cycle. By give a carefully placement of the data replicas on this two layered Chord, the less communication is needed to maintain the same target availability. The experiments show that dividing hosts by location of time zone might reduce the overhead of redundancy management.