SQL/MX 2.x Installation and Management Guide (G06.24+, H06.03+)

Measuring Performance
HP NonStop SQL/MX Installation and Management Guide—523723-004
15-7
SQL/MX Measurement Models
•
Reads
•
Records accessed
•
Records used
•
Updates
•
Writes
The FILE report can provide specific data on SQL tables. You can use the FILE report
along with other reports on a specific volume or subvolume.
NonStop SQL/MX does not collect statistics on timeouts or file-busy-time (the time
spent waiting on file I/O activities). In NonStop SQL/MX, there is no file I/O waiting as
there is for NonStop SQL/MP. The SQL/MX interface to the disk process is inherently
nowaited.
For queries that involve multiple SQL tables or compound statements, the statistics
data for the accessed tables might all be added to a single table, depending on
whether the operations are performed by the same Data Access Manager (DAM).
PROCESSH Statistics
SQLPROC and SQLSTMT are useful, but a process could be spending time in user
functions that do not contain SQL statements or in system code or system libraries. To
gather statistics on these, use PROCESSH.
The PROCESSH entity type provides information about the relative execution time of
one or more code ranges within a program. You can measure code ranges by
specifying procedure names and address ranges or by specifying a code-space
category (user code, user library, system code, or system library).
DISKFILE and DISCOPEN Statistics
The DISKFILE and DISCOPEN entities can be useful for database access and load
management. Both entities return ANSI SQL name information.
The DISKFILE entity gathers statistics for individual partitions of a file object with
multiple partitions. Use DISKFILE to determine if data is being evenly distributed
across all the partitions in a database.
The DISCOPEN entity gathers statistics for partition opens by process. Use
DISCOPEN to determine if the partitions in a database are being accessed evenly by
the processes in an application.
SQL/MX Measurement Models
When using the Measure product, you must determine whether the overhead for
gathering Measure statistics is worth the information provided by the reports. You might
find certain statistical information more meaningful with a few samplings. You should,
of course, use the Measure product for gathering detailed statistical information for
problem analysis.