1. Chapter 21: Controlling Data Storage Space 1 STAT 541 ©Spring 2012 Imelda Go, John Grego, Jennifer Lasecki and the University of South Carolina

2. 2 Outline Reducing Data Storage Space Reducing Data Storage Space Compressing Data Files Compressing Data Files Using Views to Conserve Data Storage Using Views to Conserve Data Storage There’s a level of detail here that I used to cover in STAT 740 and long ago abandoned. I’ll take the same approach for this material

3. 3 Reducing Data Storage Space Character variables Character variables –Reminder: The LENGTH statement should be listed immediately after the DATA step (even before a SET command) to take effect –Use codes rather than lengthy character variables where possible

4. 4 Reducing Data Storage Space Numeric variables Numeric variables –The LENGTH variable should only be used with integers, since it otherwise truncates significant digits from the numeric variable (sign, exponent, mantissa) –DEFAULT= assigns a default length to all subsequent numeric variables, and hence should be used with caution –PROC COMPARE can summarize rounding errors

5. 5 Compressing Data Files Uncompressed Data files have several inefficiencies Uncompressed Data files have several inefficiencies –Column space is constant for each record –Observation lengths are equal –Character variables are padded with blanks –Numeric variables are padded with 0s in the mantissa –New observations may cause an entire new page to be created

6. 6 Compressing Data Files Compressed Data files have efficiencies that you might anticipate, as well as some that would surprise you Compressed Data files have efficiencies that you might anticipate, as well as some that would surprise you –Observations are treated as a string of bytes –Blanks are removed –Consecutive repeated characters and numbers are compressed –Information on updated observations is not necessarily stored on the same page Greater overhead is required (e.g., pointers) Greater overhead is required (e.g., pointers)

7. 7 Compressing Data Files Rules for when to compress data sets are intuitive: Rules for when to compress data sets are intuitive: –Large data sets –Many long character variables –Many repeated character/numeric variables –Many missing values –Many consecutive repeated character/numeric variables

8. 8 Compressing Data Files Two options (and accompanying suboptions) for compressing files Two options (and accompanying suboptions) for compressing files OPTIONS COMPRESS=NO|YES|CHAR|BINARY OPTIONS COMPRESS=NO|YES|CHAR|BINARY –System compress (affects every data set in your SAS session) DATA dsname (COMPRESS=NO|YES|CHAR|BINARY) DATA dsname (COMPRESS=NO|YES|CHAR|BINARY) –Data set compress –YES and CHAR are good for simple character repeats –BINARY is efficient for long observations, and data with large blocks of numeric variables (e.g., testing data) –BINARY requires more CPU to uncompress SAS writes a message to LOG summarizing compression SAS writes a message to LOG summarizing compression

9. 9 Compressing Data Files By default, new observations are appended to the end of a data set (implicit OUTPUT). REUSE allows SAS to repurpose accumulated empty space in the compressed data set By default, new observations are appended to the end of a data set (implicit OUTPUT). REUSE allows SAS to repurpose accumulated empty space in the compressed data set OPTIONS REUSE=NO|YES OPTIONS REUSE=NO|YES –System reuse DATA dsname (COMPRESS=YES REUSE=YES|NO) DATA dsname (COMPRESS=YES REUSE=YES|NO) –Data set reuse Once selected, the REUSE option cannot be changed Once selected, the REUSE option cannot be changed

10. 10 Compressing Data Files Remember the use of POINT= when creating random samples in Chapter 13? This direct access has high overhead for compressed data sets and can be disabled with POINTOBS=NO to prevent this inefficient access technique Remember the use of POINT= when creating random samples in Chapter 13? This direct access has high overhead for compressed data sets and can be disabled with POINTOBS=NO to prevent this inefficient access technique

11. 11 DATA step views We introduced SQL views (partially compiled tables) in Chapter 7 as an important space-saving measure We introduced SQL views (partially compiled tables) in Chapter 7 as an important space-saving measure Views can be created in the DATA step as well (and are distinct from SQL views): Views can be created in the DATA step as well (and are distinct from SQL views): DATA dsname/VIEW=dsname; DATA VIEW=dsname; DESCRIBE;

12. 12 DATA step views Remember that views are not a panacea—they should not be called multiple times in a program since they have to read anew their source data each time the view is referenced. Remember that views are not a panacea—they should not be called multiple times in a program since they have to read anew their source data each time the view is referenced. Consider saving the view in another data set instead, then referencing that data set instead of the view Consider saving the view in another data set instead, then referencing that data set instead of the view Views must be kept current as underlying data sets change, which creates additional overhead. Views must be kept current as underlying data sets change, which creates additional overhead. Don’t create views that use files whose variable names/length/labels often change Don’t create views that use files whose variable names/length/labels often change