The section on
File System Performance Concepts discussed how distributing I/O load across different disk
drives can improve performance. Further load distribution can be
achieved by separating data and index components of SPD Server tables.
To do this, use the DATAPATH= and INDEXPATH= options when configuring
LIBNAME domains.
For example, when performing
complex WHERE clause evaluations, multiple threads are active on index
component files and the data component file at the same time. Splitting
the index and data file components onto different volumes can improve
performance by reducing disk contention and increasing the level of
parallelism down to the disk access level.
A word of caution when
using DATAPATH= and INDEXPATH= options to distribute the data and
index components: take extra care when performing and restoring disk
backups of SPD Server tables using a system backup and restore utility.
When making a backup, ensure that the metadata, data, and index component
partition files are of the same generation and are in their respective
directories.
When restoring a backup,
restore the component partitions to the same directories where they
were created. To avoid this restore problem, create symbolic links
with the original directory path that point to the restore directories.
Of course, if the components are not separated using the path options,
this restore issue does not apply.
The backup and restore
issues are not an issue when using the SPD Server Backup and Restore
Utilities. These utilities resolve any component files when backing
up or restoring tables. For more information, see Chapter 19, "SPD
Server Backup and Restore Utilities" in the
SAS Scalable
Performance Data (SPD) Server 4.5: Administrator's Guide.