NonStop Systems Introduction

NonStop Server Architecture
NonStop Systems Introduction—527825-001
7-13
Easy Expansion of Data
database. Because a database file on one disk volume can be only so large, the utility
will run up against the maximum file size limit before very long.
The SQL/MX data management systems solve the problem of rapid data expansion by
supporting a file partitioning mechanism. You can specify that a file reside entirely on
one disk volume, or that it be partitioned across several disk volumes on the same
system or on different systems. Both database files and conventional files can be
partitioned.
Each partition of a file holds a particular range of records and is physically distinct from
any other partition. However, all the partitions belong to the same logical file and
applications view a partitioned file as a single entity.
Because a file can span multiple disks, multiple I/O requests from users can be
serviced concurrently. Different disk drives perform multiple, concurrent reads and
writes of data in a single logical file. Delays caused by transaction queues are reduced,
and total system performance goes up.
By contrast, with conventional disk drives, an entire file must reside on one disk
volume, so concurrent I/O requests to the file are not possible. Each I/O request must
be queued and serviced separately. The result is slower system throughput and longer
response time.
File partitioning is also useful when multiple NonStop servers are joined in an Expand
network.
Suppose that a large library converted its entire card catalog to an online database so
that it could retrieve catalog records more quickly and maintain the catalog more easily.
The library stored this database on a single NonStop server at the main library, and the
catalog department took responsibility for maintaining the database.
Then, as the library gained additional holdings, it acquired two more buildings
elsewhere in the same city, together with a new NonStop server for each building. The
library also divided its catalog department into three branches, one in each of its
buildings, and made all three branches jointly responsible for maintaining the
database.
What will happen if the library connects its three NonStop servers in an Expand
network but leaves the entire database on the headquarters system? Every time a
library cataloger at one of the other buildings uses a workstation to access a record on
the main library system, the cataloger must wait while the request message travels
over a communications line to a server on the main library system and while the reply
from the server travels over the communications line back to the workstation.
The library can reduce network message traffic and network delays by partitioning the
database across all three systems. For example, if the primary key of the database is
the author’s last name, or it is a composite key that consists of several fields but
begins with the author’s last name, the partitions can be as follows:
•
Records for authors A-G reside on the main library (first branch) system.
•
Records for authors H-P reside on the second branch system.