Chapter 2. Infrastructure planning 25
multiple nodes writing the same file). File sizes are very large (tens to hundreds of gigabytes
per file); reuse is low, so cache algorithms are generally not helpful.
This section provides details for when the industry solution calls for a database
implementation with GPFS.
OLTP and OLAP (decision support systems)
Online transaction processing (OLTP) databases are among the most mission-critical and
widely deployed. The primary defining characteristic of OLTP systems is that the transactions
are processed in real time or online and often require immediate response back to the user.
A point of sale terminal in a retail business
An automated teller machine (ATM) that is used for bank transactions
A telemarketing site that processes sales orders and reviews the inventories
From a workload perspective, OLTP databases typically have the following characteristics:
Process a large number of concurrent user sessions.
Process a large number of transactions using simple SQL statements.
Process a single database row at a time.
Are expected to complete transactions in seconds, not minutes or hours.
OLTP systems process the daily operations of businesses and, therefore, have strict user
response and availability requirements. They also have extremely high throughput
requirements and are characterized by large numbers of database inserts and updates. They
typically serve hundreds, or even thousands, of concurrent users.
Online analytical processing (OLAP) differs from the typical transaction-oriented systems in
that they most often use data that is extracted from multiple sources for the purpose of
supporting user decision-making. The types of processing are as follows:
Data analysis applications using predefined queries
Ad hoc user queries
OLAP systems typically deal with substantially larger volumes of data than OLTP systems
because of their role in supplying users with large amounts of historical data. Whereas
100 GB of data is considered large for an OLTP environment, a large OLAP system might be
1 TB of data or more. The increased storage requirements of OLAP systems can also be
attributed to the fact that they often contain multiple, aggregated views of the same data.
Although OLTP queries are mostly related to one specific business function, OLAP queries
are often substantially more complex. The need to process large amounts of data results in
many CPU-intensive database sort and join operations. The complexity and variability of
these types of queries must be given special consideration when estimating the performance
of an OLAP system.