2.2.49 Operands related to statistical information

218) pdstbegin
Operand specification format in a HiRDB single server configuration

pdstbegin [-k statistical-information-type[,statistical-information-type]...]
      [-m output-interval-of-system-operation-statistical-information]
      [-I output-interval-of-database-I/O-statistical-information]
      [-D database-I/O-statistical-information-option
       [,database-I/O-statistical-information-option]]
      [-w]

Operand specification format in a HiRDB parallel server configuration

pdstbegin [-k statistical-information-type[,statistical-information-type]...]
      [-m output-interval-of-system-operation-statistical-information]
      [-I output-interval-of-database-I/O-statistical-information]
      [-D database-I/O-statistical-information-option
       [,database-I/O-statistical-information-option]]
      [{-x host-name | -u unit-identifier}]
      [{-a | -s server-name[,server-name]...}]
      [-w]

This operand is specified in order to begin collecting statistical information at the time HiRDB is started. The pdstend command is entered in order to stop collection of statistical information. The statistical information is output to the statistics log file.
-k statistical-information-type[,statistical-information-type]...
~<<sys>>
Specifies the type of statistical information that is to be output. The pdls -d sty command can be used to check the types of statistical information specified for output.
Statistical information type (-k option specification)Type of statistical information outputStatistical information output trigger
sysStatistical information on system operationStatistical information is output at the interval specified in the -m option.
uapStatistical information on UAPsStatistical information is output during connection to and disconnection from HiRDB.
sqlStatistical information on SQLsStatistical information is output when SQL execution starts and ends.
bufStatistical information on global buffers#1Statistical information is output at synchronization points.
filStatistical information on HiRDB files related to database manipulations
dfwStatistical information on deferred write processing
idxStatistical information on indexes
sopStatistical information on SQL static optimizationStatistical information is output when no hit occurs on an SQL object buffer during preprocessing of a dynamic SQL or static SQL.
dopStatistical information on SQL dynamic optimizationStatistical information is output when an SQL statement other than FETCH or CLOSE is executed.
pcdStatistical information related to SQL object executionStatistical information is output when an SQL object is executed.
sqhStatistical information on SQL statement statistics#2Statistical information is output during PREPARE and when an embedded SQL is executed.
objStatistical information on SQL object transmission#3Statistical information is output when an SQL object is executed.
dioStatistical information on database I/OInformation about database I/O is output at the interval (seconds) specified in the -I option.
#1: This statistical information is collected at synchronization points, and the information between synchronization points is edited. Consequently, statistical information cannot be collected unless at least two synchronization points occur. To reliably collect this statistical information, a synchronization point must be triggered by executing the pdlogswap or pdlogsync command immediately before executing the pdstend command.
#2: The statistical information on SQL statement statistics is output when the statistical information on SQLs (sql specification) is being output.
#3: This statistical information is only for a HiRDB parallel server configuration, and is not output for a HiRDB single server configuration.
-m output-interval-of-system-operation-statistical-information
~<unsigned integer> ((1-1440)) <<10>>(minutes)
This option is specified when statistical information on system operation (sys specified) is to be collected.
Specifies in minutes the interval at which the statistical information on system operation is to be output to the statistics log file.
-I output-interval-of-database-I/O-statistical-information
~<unsigned integer>((10-3600)) <<60>> (seconds)
Specify this option when statistical information on database I/O (when dio is specified for the -k option) is to be collected.
This option specifies, in seconds, the interval at which the statistical information on database I/O is to be output to the statistics log file.
If no database I/O occurs within the specified interval, no information is output to the statistics log file.
When dio is not specified for the -k option, this option is ignored.
-D database-I/O-statistical-information-option
[,database-I/O-statistical-information-option]
This option is used to control the action for measuring statistical information on database I/O. If dio is not specified for the -k option, this option is ignored.
This option can control the following items:
  • Frequency of measuring I/O times
  • Amount of statistical information that is output
The following tables show differences in actions depending on whether a value is specified for the option corresponding to each item.

Table 2-6 Table Frequency of measuring I/O times

Specified valueActionAdvantageDisadvantage
Not specifiedFor I/O that occurred on each HiRDB file comprising an RDAREA, 100 I/O times are measured per second, in the order that the input or output occurred.The impact on the performance can be minimized.For the 101th and subsequent I/O instance that occurred in a second on each HiRDB file comprising an RDAREA the I/O times are not measured.
everyioFor I/O that occurred on HiRDB files comprising RDAREAs, all the I/O times are measured.Information about all (no exception) I/O can be collected.The transaction performance deteriorates due to the overhead of measuring I/O times.
When everyio is specified, the overhead increases because all I/O times are measured for all I/O that occurred on HiRDB files comprising RDAREAs. Specify everyio only if you perform a test in advance and, based on the results, you determine that there is no problem with the performance on production operations.

Table 2-7 Table Amount of statistical information to be output

Specified valueActionAdvantageDisadvantage
Not specified
  • For each of the following items, which are collected as statistical information from the I/O information of HiRDB files comprising RDAREAs that were accessed within the output interval, information about the top 10 files in each unit is output to statistical logs:
    - Maximum number of read operations
    - Maximum read operation time
    - Total time of read operations
    - Maximum number of write operations
    - Maximum write operation time
    - Total time of write operations
  • Even if the same file becomes an output target for multiple items, the information is output without removal of duplicated data.
  • If more than 10 records have the same ordering level, the records with higher positions in the following ordering level take precedence.
    1. Ordering level defined for the RDAREA
    2. Ordering level specified for HiRDB files
Because the size of statistics logs per output trigger is fixed, the amount of used disk space and the impact on performance due to the frequent occurrence of statistics log buffer swapping are small.Because information about only the top 10 files for each unit is output, the information about all I/O for HiRDB files comprising an RDAREA cannot be collected.
everyfileThe information about all I/O for HiRDB files comprising RDAREAs that were accessed within the output interval are output.The information about all I/O for HiRDB files comprising an RDAREA can be obtained.Because the size of statistics logs per output trigger is not fixed, a large amount of disk space will be used and the frequent occurrence of statistics log buffer swapping will have a large impact on performance.
When everyfile is specified, the amount of statistics logs per output trigger is not fixed. For the pd_stj_file_size operand in the system common definition, specify the value estimated in C.1 Formulas for determining size of statistics log file (pd_stj_file_size).
For the pd_stj_buff_size operand in the system common definition, specify the value estimated using the formula provided in the description of this pdstbegin operand.
-x host-name
~<host name> ((1-32 characters))
For a HiRDB parallel server configuration, specifies a host name specified in the -x option of the pdunit operand in order to select the units for which statistical information is to be output. Specify the host name of the unit for which statistical information is to be output.
If the standby-less system switchover (effects distributed) facility is being applied to this unit, the statistical information on the guest back-end server being accepted is also output.
-u unit-identifier
~<identifier> ((4 characters))
In the case of a HiRDB parallel server configuration, specifies the unit name of a unit for which statistical information is to be output. This option is specified when it is necessary to limit the units for which statistical information is to be output.
If the standby-less system switchover (effects distributed) facility is being applied to the unit on this host, the statistical information on the guest back-end server being accepted is also output.
-a
In the case of a HiRDB parallel server configuration, specifies that statistical information is to be output for all servers. This option is specified when it is not necessary to limit the units for which statistical information is to be output.
-s server-name[,server-name]...
~<identifier> ((1-8 characters))
In the case of a HiRDB parallel server configuration, specifies the server names of the servers for which statistical information is to be output. This option is specified when it is necessary to limit the servers for which statistical information is to be output. Some types of statistical information are not output for all server types, as indicated in the following table:
Statistical information type (-k option specification)Server type
FESDSBES
sysYYY
uapY----
sqlY----
buf--YY
fil--YY
dfw--YY
idx--YY
sopY----
dopY----
pcdYYY
sqhY----
obj--YY
dio--YY
Y: Statistical information is output.
--: Statistical information is not output.
-w
Specifies that statistical information about the thread-to-thread lock-release wait time is to be output. This option is enabled when statistical information about system activities is output. Therefore, statistical information about the thread-to-thread lock-release wait time can be acquired if the -k option is omitted or the specified -k option contains sys.
This option might affect the performance of the entire system; therefore, we recommend that you normally do not specify this option.
Specification guideline
  • If A is specified for the pd_statistics operand or is assumed to be the default value (recommended mode), you do not need to specify the -w option. The following shows the statistical information to be collected when the pd_statistics operand is A:
    - Statistical information on system operation (one-minute interval)
    - Statistical information on global buffer pools
    - Statistical information on deferred write processing
    - Statistical information on database I/O
    To collect statistical information other than the above, we recommend that you specify both the above statistical information and the additional statistical information to be collected.
    Specification example:
    The following is an example of specifying collection of additional statistical information on UAPs:
    pdstbegin -k sys,buf,dfw,dio,uap -m 1
  • If Y or N is specified for the pd_statistics operand or if N is assumed to be the default value (compatibility mode), we recommend that you specify settings to collect the following statistical information:
    - Statistical information on system operation (one-minute interval)
    - Statistical information on global buffer pools
    - Statistical information on deferred write processing
    - Statistical information on database I/O
    Specification example:
    pdstbegin -k sys,buf,dfw,dio -m 1
Relationship to other operands
  • This operand is related to pd_stj_file_size.
  • When this operand is specified, specification of the pd_statistics operand is ignored.
Notes
  • The pdstbegin operand can be specified only once. If it is specified more than once, the first time it is specified is valid and the subsequent specifications are ignored.
  • When the pdstbegin operand is specified, the specified statistical information is output until HiRDB is terminated or until the pdstend command is entered.
  • If you use the standby-less system switchover (1:1) facility, the specification for this operand is invalid when the system is being switched to an alternate BES unit.
  • Note the following if you use the standby-less system switchover (effects distributed) facility: During a normal HiRDB startup, each server follows the value specified for this operand. During a HiRDB restart (including a system switchover), the value specified for this operand is invalid. During a restart, the previous statistical information collection state is inherited. During a system switchover, the statistical information prior to the switchover is inherited.
  • If no server has started in the unit, unit statistical information is not acquired.
  • When Real Time SAN Replication based on the log-only synchronous method is used, statistical information cannot be acquired at the log application site. If this operand is specified, the KFPS04689-W message is displayed.
  • Depending on when the pdstbegin and pdstend commands are entered, the statistical information on UAPs might not match the statistical information on SQLs. The relationship between when the pdstbegin and pdstend commands are entered and the statistical information that is output is shown as follows:

    [Figure]

  • If the HiRDB system, unit, or server is terminated (including abnormal termination) and is then started again, the statistical information collection state might not be inherited. By specifying the pd_statistics operand or this operand, statistical information can continue to be collected during a restart or a system switchover. The following table shows whether the statistical information collection state is inherited when the HiRDB system, unit, or server is started.
    Start modeStatistical information collection environmentStart condition
    HiRDB
    start
    Unit start (start of single unit)Server start (start of single server)
    Standby-less system switchover (effects distributed) facility is not usedStandby-less system switchover (effects distributed) facility is used
    Normal startStatistical information is being collected by specifying the pdstbegin or pd_statistics operand.YYN#Y
    Statistical information is being collected by executing the pdstbegin command.N#N#N#N#
    RestartStatistical information is being collected by specifying the pdstbegin or pd_statistics operand.YY--I
    Statistical information is being collected by executing the pdstbegin command.N#N#--Y
Y: Statistical information collection state is inherited.
I: Statistical information collection state is inherited. When the pdstbegin command is executed, the collection state of the statistical information specified by this command is inherited.
N: The statistical information collection state is not inherited.
--: Not applicable.
#: To collect statistical information, you must execute the pdstbegin command after starting the HiRDB system.
219) pdhibegin -k statistics-type[,statistics-type]...
Specifies the types of statistical information to be collected beginning at the time of HiRDB startup. When this operand is specified, statistical information is output continuously to the system log file until HiRDB is terminated.
-k statistics-type
Specifies the type of statistical information to be output:
cnc: Statistical information related to CONNECT/DISCONNECT.
In the case of a HiRDB parallel server configuration, the output destination file is the system log file at a front-end server.