Databases Reference
In-Depth Information
Data is distributed within a cluster of autonomous disks, and can be uniformly
accessed from multiple hosts simultaneously. The disks are also able to handle
data distribution, load skews and disk failures or errors in the disk controllers
autonomously, and can reconfigure clusters after a repair. These aspects are
accomplished within the cluster of autonomous disks without interaction with the
hosts. This provides excellent transparency to the hosts.
To implement autonomous disks, we utilize a distributed directory effectively,
and define three command layers: external interface stream (EIS) commands,
internal stream (IS) commands, and internal disk access (IDA) commands. The
EIS commands are implemented using the IS commands and ECA rules. The
combination of command layers and rules enables users to describe the behavior
of the autonomous disks easily. Moreover, the combination also provides flexibility
in varying management strategies for individual streams. Thus, hosts are not
involved in communication between disks to realize these functions. This provides
high scalability of the system.
The transparency, flexibility, and scalability derived from this approach are
essential. The variety of functions provided for hosts and disks allows the
applications on the hosts to be independent of the disk configuration. Autonomous
disks are effective in a scalable network environment containing a number of
hosts and disks.
By estimating operation costs in autonomous and ordinary disks, we find that
the autonomous disks reduce data transfer overheads compared to the same
operations on ordinary disks. This indicates that the autonomous disks are suited
to the network environment in respect to both functionality and performance.
Because we treat a file as a stream in the autonomous disks they are especially
applicable to many advanced applications. For instance, Internet providers or web
directory servers having a large number of HTML files have difficulty in maintaining
files. Business-to-business e-commerce frameworks also generate enormous
numbers of XML files. The proposed autonomous disks are capable of handling
file streams very effectively for such situations, using the distributed directory. We
can also provide a dedicated search EIS command to search such files for required
components.
These functions can be implemented in PC clusters, although we intend to install
them eventually in functional disks having disk-resident processors. We are currently
implementing an experimental system using PCs connected by a network to evaluate
the mechanism.
Bibliography
1)
Jim Gray and Andreas Reuter.
Transaction Processing: Concepts and Techniques.
Morgan Kauf-Mann, 1993.
2)
David A Patterson, Garth Gibson, and Randy H.Katz. A Case for Redundat Arrays of
Inexpensive Disks(RAID). In
Proc. of ACM SIGMOD Conference,
pages 109-116,
Jun 1988.
