Increasing Storage Performance and Reliability with Magnetic RAM

University Of California-Santa Cruz, Santa Cruz CA

Investigators

Abstract

The research dramatically improves file system performance and reliability by using Magnetic RAM as a permanent store for both file system indexes (metadata) and data. Keeping metadata in MRAM increases the speed of metadata operations by more than an order of magnitude while decreasing software complexity and improving integrity. This research also investigates dynamic memory-based algorithms for metadata management, reducing metadata size and complexity while increasing metadata flexibility. The research explores the effects of varying MRAM sizes and performance characteristics on file system performance and the choice of file system algorithms. The higher file system performance and higher reliability made possible by this research has several far-reaching impacts. First, file system performance improvements of an order of magnitude dramatically reduces file server costs, making storage more affordable. Second, the reduced code complexity enabled by the use of memory-based algorithms permits faster development of simpler file servers. Third, the increased reliability from lower code complexity and continuous integrity checking allows the production of much more reliable file servers. The research is also applicable to systems built with other emerging non-volatile memory technologies such as FeRAM and new types of Flash RAM.