DTI: 2010–11 Funded Proposals

David J. Lilja

GSA: A New Storage Architecture Combining Graphic Processing Units and Solid State Disks to Boost I/O Performance

While storage capacity and processing power have experienced rapid growth in the past, improvement in data bandwidth and access times of data storage devices have not kept pace. As a result, we are seeing an ever widening speed gap between processing elements and I/O storage systems. This proposal aims at minimizing this speed gap by proposing a new storage architecture named GSA to leverage the high-performance computation available from graphic processing units (GPUs) and the fast read performance of solid state disks (SSDs). If successful, we can expect an order-of-magnitude performance speedup using GSA over traditional storage architectures. In order to investigate whether GSA can deliver such high performance, we will conduct extensive architectural research from the following four aspects: (1) studying temporal and content localities for various applications and benchmark programs and exploiting these two localities to speed up I/O performance, (2) designing an efficient GPU-based similarity detection algorithm for discovering content locality, (3) looking for an optimized way to layout data on SSDs and HDDs, and (4) developing an optimal scheduling algorithm and metadata structure for this novel GSA storage architecture.