Michael Reiter, University of North Carolina - August 2014

Hybrid clouds, involving storage and computation within private datacenters regularly but at public data centers when opportune, offer a potentially flexible means to leverage public clouds on an as-needed basis. This theoretical flexibility can be hampered, however, by the various costs associated with migrating data and computation between private and public data-centers  whether that be in terms of task latency, inter-datacenter bandwidth consumed, capacity limitations associated with datacenters, monetary costs for storing data in public clouds, etc. We propose to extend a scalable object store called WACCO, originally built to leverage object migration among data centers to minimize access latency on a per-object basis, to experiment with algorithms for managing migration so as to account for migration, access, and storage costs in aggregate, across objects. We plan to develop migration algorithms for this goal and demonstrate their improvements over existing migration in WACCO through trace-driven experiments leveraging multiple real workloads.