Model Driven Approach for Distributed Storage Management:

1) A Generic Policy Management framework that will enable to define storage permissions, policies
(for placement, caching, replication, migration, weaver-erasure, bandwidth allocation strategies).
A Policy will be a set of Rules with precedence and interdependencies (on other policies).

2) A Generic Workflow Engine that will define various workflows with tasks (having triggers, timers)
corresponding to the various strategies. For example some workflows could approximately be like:

FT-WF: Fault Tolerance where a depot failure triggers a workflow engine to execute a certain
predefined workflow of different tasks to be performed. When primary Depot fails, 1. fallback to
secondary which has the most optimal location, 2. initiate state and data migration task,
3. initiate the necessary replication tasks, 4. notify the metadata server about the new depot in
order to updates the exnodes properly.

Rep-WF: Replication that determines the degree and where to replicate. 1. Choose optimal depot
locations for replication 2. Spawn n replication tasks where n=degree

of replication 3. Copy data
between primary and secondary depots

3) Workflows will have tasks that may trigger additional workflows. For example the FT-WF has the
migration and replication tasks that may trigger their corresponding workflows (Rep-WF).

4) Depot Data Store Placement based upon global PDF of the clients as well based upon context (place
a depot in the locality that has the maximal number of clients whose data is stored in it)

Topics for Papers to read: GPFS, CEPH, IBP, IROD, FAST-TCP, CDN, DFS, GridFTP

Symposiums: SAC ‘08
Conferences: InfoScale ‘08
Workshops: UPGRADE ‘08, MIR ‘08