Previous Page | Next Page

SAS Data Set Options

OBSBUF= Data Set Option



Determines the size of the view buffer for processing a DATA step view.
Valid in: DATA step and PROC steps
Category: Data Set Control
Restriction: Valid only for a DATA step view

Syntax
Syntax Description
Details
Example
See Also

Syntax

OBSBUF=n


Syntax Description

n

specifies the number of observations that are read into the view buffer at a time.

Default: 32K bytes of memory are allocated for the default view buffer, which means that the default number of observations that can be read into the view buffer at one time depends on the observation length. Therefore, the default is the number of observations that can fit into 32K bytes. If the observation length is larger than 32K, then only one observation can be read into the buffer at a time.
Tip: To determine the observation length, which is its size in bytes, use PROC CONTENTS for the DATA step view.
CAUTION:
The maximum value for the OBSBUF= option depends on the amount of available memory.

If you specify a value so large that the memory allocation of the view buffer fails, an out-of- memory error results. If you specify a value that is larger than the amount of available real memory and your operating environment allows SAS to perform the allocation using virtual memory, the result can be a decrease in performance due to excessive paging.  [cautionend]


Details

The OBSBUF= data set option specifies the number of observations that can be read into the view buffer at a time. The view buffer is a segment of memory that is allocated to hold output observations that are generated from a DATA step view. The size of the buffer determines how much data can be held in memory at one time. OBSBUF= enables you to tune the performance of reading data from a DATA step view.

The view buffer is shared between the request that opens the DATA step view (for example, a SAS procedure) and the DATA step view itself. Two computer tasks coordinate between requesting data and generating and returning the data as follows:

  1. When a request task, such as a PRINT procedure, requests data, task switching occurs from the request task to the view task in order to execute the DATA step view and generate the observations. The DATA step view fills the view buffer with as many observations as possible.

  2. When the view buffer is full, task switching occurs from the view task back to the request task in order to return the requested data. The observations are cleared from the view buffer.

The size of the view buffer determines how many generated observations can be held. The number of generated observations then determines how many times the computer must switch between the request task and the view task. For example, OBSBUF=1 results in task switching for each observation, while OBSBUF=10 results in 10 observations being read into the view buffer at a time. The larger the view buffer is, the less task switching is needed to process a DATA step view, which can speed up execution time.

To improve efficiency, first determine how many observations fits into the default buffer size, then set the view buffer so that it can hold more generated observations.

Note:   Using OBSBUF= can improve processing efficiency by reducing task switching. However, the larger the view buffer size, the more time it takes to fill. This delays the task switching from the view task back to the request task in order to return the requested data. The delay is more apparent in interactive applications. For example, when you use the Viewtable window, the larger the view buffer, the longer it takes to display the requested observations, because the view buffer must be filled before even one observation is returned to the Viewtable. Therefore, before you set a very large view buffer size, consider the type of application that you are using to process the DATA step view as well as the amount of memory that you have available.  [cautionend]


Example

For this example, the observation length is 10K, which means that the default view buffer size, which is 32K, would result in three observations at a time to be read into the view buffer. The default view buffer size causes the execution time to be slower, because the computer must do task switching for every three observations that are generated.

To improve performance, the OBSBUF= data set option is set to 100, which causes one hundred observations at a time to be read into the view buffer and reduces task switching in order to process the DATA step view with the PRINT procedure:

data testview / view=testview;
   ... more SAS statements ...
run;

proc print data=testview (obsbuf=100);
run;


See Also

Data Set Options:

SPILL= Data Set Option

Previous Page | Next Page | Top of Page