|
When a node in the system wants to read a data-item it accesses the closest
replica of the data-item. However, updates of a data-item are always initiated at
root of the tree. A node that wants to update a data-item, sends it to the root,
which is then propagated down the tree-hierarchy. The root node as well as the
other nodes holding replicas collect various statistics such as the frequency at
which a replica is used, frequency at which the data-item is updated, ratio of
reads to writes for a data item, failure statistics of the node holding the replica,
and storage availability and utilization at each node. The office network configuration information collected
at various nodes will be aggregated at the root of the tree. These statistics
will be fed into our optimization engine which produces solutions to the replica
placement problem, indicating which nodes should hold additional replicas and
which of the existing replicas need to be de-commissioned. The replication tree
is then modified accordingly.
|
