In the SAS® System, SAS I/O, the most common input/output (I/O) processing, is performed on SAS files such as SAS data sets, 
SAS catalogs, and SAS index files. Most SAS I/O processing is sequential I/O, which means that the DATA step and SAS procedures read a data set from the beginning of the file to the end of the file. The data is read from or written to pages that are typically 4—64 K in size.

A SAS job can contain numerous DATA steps and SAS procedures that read and process the same data set or 
distinct data sets, or it can write or rewrite numerous new and existing data sets. With standard SAS 
options, each data set that is read from disk is read into the Microsoft Windows file cache. 

Likewise, before a data set is written to disk, it is written to the Windows file cache.
A large SAS job can cause numerous SAS data sets to reside in the Windows file cache. However, 
processing large SAS data sets (data sets greater than 2 GB) reduces I/O throughput. This loss of I/O
throughput is not obvious typically because data sets tend to slowly increase in size. Eventually, the
processing of such a job seems much slower than is expected, but you might not recognize that reduction
in I/O throughput is the root cause. Often, customers spend many hours investigating the hardware of the
system by running the SAS job as well as upgrading or enhancing the system, only to learn that none of
these changes solve the problem.

This paper explains how experienced users can recognize any loss of I/O throughput before they resort to
hardware or system changes and how they can solve this problem by using the SGIO (scatter-read/gather-
write input/output) feature.