HP Integrity rx1620 Server 1.60 GHz 267 MHz FSB Base System AB431A Dépliant
Codici prodotto
AB431A
Increased storage requirements
To support the commit/rollback protocol, RDBMSs maintain a transaction log that will allow for the
To support the commit/rollback protocol, RDBMSs maintain a transaction log that will allow for the
potential rollback of non-committed updates. To optimize precision searches for specific applications,
RDBMSs support the creation of special access data structures, such as b-tree indices and hash tables.
RDBMSs support the creation of special access data structures, such as b-tree indices and hash tables.
Some RDBMSs support isolation concurrency through the storage of lock information along with the
data. To support data versioning, multiple versions of the data are stored.
This is just a partial list of the RDBMS features that result in increased storage requirements. The net
result is the volume of RDBMS storage overhead may be up to three times larger than that of the
This is just a partial list of the RDBMS features that result in increased storage requirements. The net
result is the volume of RDBMS storage overhead may be up to three times larger than that of the
original data, resulting in a four to one expansion of storage requirements.
Increased CPU requirements
Row-level concurrency control requires that RDBMS check a lock/transaction table for every row
Row-level concurrency control requires that RDBMS check a lock/transaction table for every row
access. Most RDBMSs use a paging system to allow multiple transactions to share data and reduce
I/O requirements. These and other internal structures use additional CPU cycles.
Increased I/O requirements
Increases storage requirements result in increased I/O requirements. Indices must me maintained in
Increases storage requirements result in increased I/O requirements. Indices must me maintained in
tandem with every row update.
Geometrically decreasing load performance
As the volume of data increases, the cost and time to load the next row of data increases
As the volume of data increases, the cost and time to load the next row of data increases
geometrically. This is due to the maintaining of indices. RDBMSs are optimized for precision
searching of data, which loads one transaction at a time; and not meant for bulk data loading. Due
to RDBMS indices, this loading-rate degradation is permanent. Although the absolute numbers may
vary based on hardware configuration, the shape of the curve is the same.
Search specific optimization
Search is optimized in RDBMSs by creating indices which anticipate a specific set of search criteria.
Search is optimized in RDBMSs by creating indices which anticipate a specific set of search criteria.
For example, if it is anticipated that many searches will involve a user name, then an index will be
created on the user name column. Searches which fall within the bounds of the anticipated search
criteria will execute quickly. Searches that do not fall within the bounds of the anticipated criteria will
cause a complete table scan.
Some common event data searches cannot be optimized by an RDBMS at all. These include substring
cause a complete table scan.
Some common event data searches cannot be optimized by an RDBMS at all. These include substring
and pattern-matching searches. These also result in a complete table scan. For large databases with
billions of rows, these searches can become multi-day queries rendering a result of no practical value.
Example: A customer attempted to perform a search against one billion web proxy events that
would return timestamp, IP address, user ID and URL for all records where the URL contained a
specific string. As this is a pattern search, the RDBMS was required to conduct a full table scan
and could take advantage of any indices. This resulted in a search that took days to complete
using high-cost, massive multi-processor, and high-performance servers.
Search optimization and load rate trade-off
To optimize a greater number of search scenarios, more indices must be created. The more indices
To optimize a greater number of search scenarios, more indices must be created. The more indices
that are created, the slower the event data load rate becomes. This forces the database administrator
to constantly maintain a balance between search performance and load performance. To do so,
indices have to be created and dropped. Consequently, search scenarios must be anticipated and
prioritized. If the event data load rate becomes unacceptable, then indices must be dropped.
prioritized. If the event data load rate becomes unacceptable, then indices must be dropped.
Dropping an index can reduce search performance in ways that are hard to predict. For large
volumes of event data, creating and dropping indices are a major undertaking that can temporarily
shut down database operations and take hours or days to complete.
8