CSR:Small:RUI: A Fast Backing Store System on top of Network RAM, Flash, and other Cluster-wide Storage
Donor: National Science Foundation
Award Date: 8/27/2011
It is increasingly important to design parallel computing systems that efficiently support data-intensive computing. Parallel scientific applications, such as global climate modeling, often perform large amounts of disk Input/Output because the data they process cannot all fit at one time into physical computer memory (RAM). As the disparity continues to grow between the speeds of magnetic disk and other hardware such as RAM, interconnection networks, and solid state devices, the cost of disk I/O is increasingly the bottleneck of system performance. Recent advances in fast random access storage devices means not only that magnetic disk will likely become obsolete in the near future, but also that backing storage in parallel systems will be more heterogeneous, complicating future parallel operating system design.
The objective of this project is to solve problems associated with supporting heterogeneous backing storage in parallel systems. The goals are: to provide parallel cluster operating systems with a simple interface to cluster-wide storage, freeing them from needing specialized policies for every combination of backing storage device; to create a cluster-wide backing store system that adapts to changes in cluster-wide resource utilization and storage capacity; and to take advantage of the strengths of different underlying physical storage devices.
Project Period: 9/1/2011 – 9/30/2016