A fundamental limit has been reached in computation systems with a monolithic magnetic disk storage. With the advent of Big Data computations, the traditional storage is a bottleneck in a computation process that involves the computation system’s CPU, memory, and storage; no longer is the CPU and the memory the biggest bottleneck in Big Data computations. The Computation Structures Group has implemented a computation system with a new storage system: a distributed flash disk storage. In this new computation system, there is a storage-controller hardware that is re-configurable (via hardware-description software), offering a degree of freedom to exploit in Big Data computations. My specific role in this project is to implement, in hardware-description software, those reconfigurations needed in MapReduce computations.

My background is a combination of two courses relevant to this project: 6.111 which taught me the essence of re-configurable hardware and 6.033 which introduced MapReduce, a popular venue of Big Data computations.