The HPSUMMARY procedure calculates descriptive statistics and is available with the SAS^® High-Performance Analytics product. An introduction to PROC HPSUMMARY can be found in the HPSUMMARY Procedure documentation.

The HPSUMMARY procedure is multi-threaded and can operate in either distributed mode on a cluster of networked computers, or in single-machine mode. When running in either mode, the HPSUMMARY procedure is limited to operating within memory and, unlike its Base SAS^® counterpart, PROC SUMMARY, does not use utility files when memory limits are exceeded. Because of the difference in intended operating environments, processing capabilities, and performance, you should consider the following best practice advice when choosing between PROC HPSUMMARY or PROC SUMMARY:

• PROC HPSUMMARY run in distributed mode does not perform well with high cardinality class variables. It might require longer processing time to compute statistics for a large number of class levels and numerous class types because network exchange of data occurs, and network communication can be slow with large amounts of data.
• PROC HPSUMMARY performs best in distributed mode when the cardinality of the class variable(s) is low.
• When running in single-machine mode, the processing of PROC HPSUMMARY is limited by the memory available on that machine and might fail when computing statistics for a large number of class levels. In this case, PROC SUMMARY is a better choice because it uses a utility file to complete processing when a memory limit is reached.