Database Reference
In-Depth Information
in many projects where storage space management and dynamic allocation of
storage is not a concern.
Another approach to space allocation is to provide a software layer that
can operate directly on a variety of hardware, such as a raw disk or a RAID
(redundant array of independent disks) system, or directly on a local file sys-
tem. This approach was taken by the network storage (NeST) project,
2
which
provides an abstract interface to disks. The idea is to have a relatively simple
software layer that exposes a storage appliance interface. NeST provides dy-
namic storage allocation in a form of “lots” that can be allocated to a client.
It can manage guaranteed and “best effort” type lots and provides multiple
protocols for file access. However, unlike SRMs, NeST is not designed to inter-
act with mass storage systems (MSSs) or complex storage systems that have
multiple components.
3.2 Support for Mass Storage Systems
3.2.1 Management of Files in Today's
Mass Storage Systems
Today's MSSs, including the largest file and archival storage systems, are
designed with high performance and reliability as their key design principles.
The main goal for performance is the rate at which a user can read or write
data to or from the system. Reliability is the degree to which users trust data
that exists in the system through many decades and technology changes. In
addition, most large file and archival storage systems adhere to standards that
contribute to their portability but ultimately limit their features down to a
set of known commands or functionality. For instance, many file or archival
storage systems comply with the portable operating system interface for unix
(POSIX) standard for I/O and provide very few special or custom commands
outside the standard set of commands for common Unix file systems.
Given the constraints of designing for reliability as opposed to performance
and self-imposed limitations in favor of portability, MSSs do not compete with
the data performance of high-performance file systems. High-performance file
systems make design and implementation trade-offs that would affect reliabil-
ity or not be acceptable to an MSS in order to achieve the higher-performance
data rates. An acceptable target for one commercial MSS is to stay within 10%
or an order of magnitude of what a file system is capable of handling in terms
of data rates. Depending on their ability to stripe transfers across multiple
disk or tape devices, most MSSs are capable of extremely high data rates which
are comparable to high-performance file system data transfer capabilities.
Another trend in MSS evolution is the average size of user files. The av-
erage size of files has not appreciably increased over the past few decades
