| Next Page | Prev Page | Main Index | Home |
MPbase often makes it possible to attain "hard-disk-like performance" with the use of less costly removable storage media. This allows the optional use of optical or, in some cases, even tape as the storage media for a database.
There are several aspects of MPbase that make it ideal
for any database or data warehouse using fixed, removable, or
mixed media. First, the process of naturalizing the data creates
a segmented sequential internal layout. Second, by working with
the file system directories, use of cache is maximized. Third,
due to the parallel nature of MPbase the effects of wait
times are minimized. Fourth, the compression both reduces the
amount of data transferred and increases the speed at which the
search executes.
The process of naturalizing data creates a productive layout for
removable media. MPbase breaks the information up into
lots of discrete files. These files are in groups such that one
or two large blocks will tend to resolve most queries. The placement
of data in the naturalized format tends to minimize the number
of such blocks needed by any one query. And, MPbase at
the same time reduces the number of disks or tapes needed by any
single query.
The indexing is to a large degree resolved by the file system
index. This index is almost always in a memory cache and is at
least partially loaded during a mount. In the case of media management
libraries, this information is normally kept on disk. This allows
a very quick selection of the sections of the database that will
need to be searched.
As the MPbase is massively parallel it can be working on
many queries or sub-queries at the same time. This reduces the
impact of media waits. With many parallel internal paths the searching
of blocks already in memory occurs in parallel with the media
waits for new blocks.
When added to the compression, which makes the system more CPU
rather than I/O-bound, you create the best possible scenario for
a database containing removable media. Keep in mind that this
system is NOT a waster of CPU time. It reads the data smart, not
hard. The structure of the naturalized data allows it to spend
most of its time on useful work rather than on reading and parsing
unnecessary data blocks.
Although each of the above will help with removable media, all of the above have a tremendous effect on the total throughput.
| Next Page | Prev Page | Main Index | Home | © 1998-2004 NPSI |