Swift: A Storage Architecture for Large Objects

Abstract

Managing large objects with high data-rate requirements is difficult for current computing systems. We describe an Input/Output architecture, called Swift, that addresses the problem of storing and retrieving very large data objects from slow secondary storage at very high data-rates. Applications that require this capability are poorly supported in current systems, even though they are made possible by high-speed networks. These range from storage and visualization of scientific computations to recording and play-back of color video in real-time. Swift addresses the problem of providing the data rates required by digital video by exploiting the available interconnection capacity and by using several slower storage devices in parallel. We have done two studies to validate the Swift architecture: a simulation study and an Ethernet-based proof-of-concept implementation. Both studies indicate that the aggregation principle proposed in Swift can yield very high data-rates. We present a brief summary of these studies.