Disk Array Architectures
Walt Burkhard, UC San Diego
Abstract
A variety of applications, such as on-line transaction
processing and scientific computing, require high throughput from
cost-effective storage subsystems. Disk arrays offer significant
advantages over individual disks. By aggregating multiple disk drives,
arrays achieve better transfer bandwidths and potential reductions on
seek and rotational latencies due to smaller-sized form factors. Reliability
is obtained by storing redundant data. Data is grouped into stripes
that contains redundant data as well. If a disk is unavailable,
all the stripe units mapped to the disk can be recovered by virtue of
the redundant data.
In declustered data organizations, each stripe is mapped to k of the
n disks thereby achieving significant performance improvements during
degraded operation. These layouts are based upon balanced incomplete
block designs and parity declustering is an example. We will present
several alternative approaches that provide very attractive implementations
as well as excellent run-time performance.
Biography
1972-present.
Department of Computer Science and Engineering
UC San Diego
1969-1972.
Department of Electrical Engineering
Princeton University
Ph.D. UC Berkeley 1969