CISE Research Resources: Instrumentation Support for Very Large Data Stores

点击翻译按钮获取中文摘要

EIA 02-24439Grunwald, Dirk C.University of ColoradoTitle:CISE RR: Instrumentation Support for Very Large Data Stores This proposal, designing a system using Massive Arrays of Idle Disks (MAID) as a replacement for data centers that would normally use tape archives, will utilize infrastructure in the design and fabrication of an alternative design to conventional mass storage systems that attempt to address critical data center issues such as data density (i.e., reduced floor space), reliability, power efficiency, and manageability. Using a combination of commodity disk drives, power management, distributed control, and caching hierarchy, the researcher claims that the MAID storage organization provides storage densities matching or exceeding those of tape libraries with performances similar to disk arrays. Studies have shown that through a combination of effective power management of individual drives and caching, this performance can be achieved using a very small power envelope. The large backing store will be composed of inexpensive IDE disks that, rather than spinning, are allowed to go idle when not in use. The main goal consists of replacing conventional tape archival storage with an alternative technology at significantly reduced power cost. The experimental testbed consists of 120 IDE drives, each of which holds 160GB of data for a total of 19.2TB of online storage, of which approximately 160TB would be available once redundancy is addressed. The infrastructure also includes 120 drive enclosures that connect commodity IDE drives to 1394 (firewire) I/O networks.