This project takes a fresh look at the building blocks of high-end storage systems—individual servers in a complex that typically scales to thousands of nodes—in a concentrated effort to accelerate I/O capacity with advanced, yet affordable, storage and I/O subsystems. Combining advanced chip sets that feature fast, wide buses on the memory controller with commodity RAID controllers yields systems with local storage bandwidth measured in multiple GBps.

Building a networked storage system on this hardware yields mixed results: preliminary experiments show that operating systems must be re-engineered to scale to this level of performance. We also explore the use of enhanced I/O in a well-understood computational problem: solving boundary value problems with numerical methods.

What you will learn:

• Engineering considerations in the design of storage systems
• Operating system limitations in high-end networked storage systems
• How a university research lab approaches innovation through experimental research