|
Computer systems have long been faced with problems of scale and performance.
Many approaches have been taken across a broad spectrum of computing resources,
software environments and workloads. We are developing a storage system for
large-scale distributed data-mining systems, which are faced with two particular
problems that go above and beyond traditional systems: an overabundance of
potentially valuable data (more than they could possibly store regardless of
capacity) and an enormous quantity of extremely short-lived data (much of it
in small individual objects, which would stress traditional file systems).
VBR allows a storage system to determine automatically which information to
store based on the relative valuation of all of the data in the system and to
recycle the space occupied by deleted data extremely efficiently. Rather than
deleting data manually or after a fixed time period, the system deletes the
lowest valued data. These values can change automatically, for example,
declining linearly over a period of time. We are also exploring the extension
of this approach to other time-varying parameters, such as performance
objectives.
IBM Almaden Research - Storage Management and Solutions
|
Time-varying Management of Data Storage
Ranjita Bhagwan, Fred Douglis, Kirsten Hildrum, Jeffrey O. Kephart, and William E. Walsh;
First Workshop on Hot Topics in System Dependability, June 2005.
Position: Short Object Lifetimes Require a Delete-Optimized Storage System
Fred Douglis, John Palmer, Elizabeth S. Richards, David Tao, William H. Tetzlaff, John M. Tracey, and Jian Yin;
11th ACM SIGOPS European Workshop, September 2004.
|
