Processor performance has increased at the phenomenal rate of 60 percent per year (Moore's Law), while disk access time, which is limited by mechanical delays, has improved by only about 8 percent per year. Further compounding the widening gap between processor and disk storage performance is the fact that disk capacity has doubled almost every year. As a result, it is increasingly difficult to use all of the available disk capacity effectively. Note, however, that although disk access time has been relatively stable, disk transfer rate has been increasing by about 40 percent per year
In this research, we are exploring ways to exploit increasing processor performance, disk capacity, and sequential bandwidth to mask the growing performance gap between processor and disk storage. We have considered embedding processing power near the data (for example, on the disk adapter) to handle general-purpose processing offloaded from the host system. Our current interest is in reorganizing data to increase spatial locality, thereby improving disk performance. Our methodology is trace-driven simulation, and an ongoing part of this research is the collection and characterization of high-quality I/O traces.
| 
