The pdstbegin command outputs to the statistics log file statistical information about a unit located at the server machine with a specified host name. When output of statistical information is to be stopped, the pdstend command must be executed.
The statistical information is output to the statistics log file named pdstj1 or pdstj2 at each host.
The obtained statistical information can be read by the statistics analysis utility. For details about the statistics analysis utility, see Chapter 14. Statistics Analysis Utility (pdstedit).
HiRDB administrator
pdstbegin [-k statistical-information-type[,statistical-information-type]...] [-m system-operation-statistical-information-output-interval] |
pdstbegin [-k statistical-information-type[,statistical-information-type]...] [-m system-operation-statistical-information-output-interval] |
Specifies the type of statistical information to be output:
When system activity statistical information (-k sys specified) is to be output to the statistics log file, specifies in minutes the interval at which the information is to be output.
When statistical information about database I/O operations is to be output to the statistics log file (-k dio is specified), this option specifies, in seconds, the interval at which the information is to be output.
If no database I/O operations occur within the specified interval, there is no output to the statistics log file.
When -k dio is not specified, this option is ignored.
If the pdstbegin -I command is executed while database I/O statistical information is being output, the time at which the command is executed is set as the start point for the interval. The following figure shows the interval at which information is output.
Figure 2-14 Interval specified in the -I option
As the statistical information output interval becomes shorter, the volume of statistics log information that is output increases and the frequency of statistics log buffer swapping increases, resulting in an increase in the system's workload. For this reason, change the default output interval to a smaller value only when detailed investigation is needed. Once the needed information has been acquired, change the setting back to the previous value.
Specifies management of the measuring operation for statistical information about database I/O operations. This option is ignored when -k dio is not specified.
This option can manage the following items:
The following tables describe the operational differences depending on the values that are specified for these items.
Table 2-33 Frequency of I/O time measurement
Value | Action | Advantage | Disadvantage |
---|---|---|---|
Omitted | Measures the input/output times for 100 I/O operations per second in the order of the operations that occur on each HiRDB file constituting an RDAREA. | Adverse effects on performance can be minimized. | The input/output times for I/O operations after the first 100 I/O operations that occur within the measurement second on each HiRDB file constituting an RDAREA are not measured. |
everyio | Measures the input/output times of all I/O operations that occur on HiRDB files constituting an RDAREA. | All input and output information can be acquired. | Transaction performance is affected adversely because of the overhead of measuring input/output times. |
When everyio is specified, overhead increases because the input/output time is measured for every I/O operation on an RDAREA that is processed by the transaction. Perform a test in advance in which you specify this option only to determine if it will have adverse effects on performance during actual application processing.
Table 2-34 Amount of statistical information that is output
Value | Action | Benefit | Disadvantage |
---|---|---|---|
Omitted |
| Because the size of the statistics log information that is output at one time is fixed, the required disk space is small, thereby minimizing adverse effects that result from frequent statistics log buffer swapping. | Because information about only the top 10 files is output for each unit, input/output information for all HiRDB files that constitute an RDAREA might not be obtained. |
everyfile | Input/output information is output for all HiRDB files that constitute an RDAREA that was accessed within the output interval. | Input/output trends for all files constituting an RDAREA can be output. | Because the size of the statistics log information that is output at one time is not fixed, a large amount of disk space is used, thereby increasing the adverse effects of frequent buffer swapping. |
When everyfile is specified, the amount of statistics log information that is output at one time is not fixed. Specify in the pd_stj_file_size operand in the system common definition the value obtained from Formulas for determining size of statistics log file (pd_stj_file_size) in the manual HiRDB Version 9 System Definition.
Specify in the pd_stj_buff_size operand in the system common definition the value obtained from the estimation formula shown in pd_stj_buff_size in the manual HiRDB Version 9 System Definition.
To change the measuring operation while output of statistical information about database I/O operations is in effect, execute the pdstbegin -k dio command with the -D option specified. When you do this, the default operation is applied to the items that are not specified in the -D option.
If the -D option is omitted, the current operation is inherited.
The example shown in the following figure specifies the -D option to change the measuring operation.
Figure 2-15 I/O statistical information measuring operation specified with the -D option
Table 2-35 Measuring operation for each interval
Interval | Operation for each item | |
---|---|---|
Frequency of I/O time measurement | Amount of statistical information that is output | |
A | Not specified | Not specified |
B | everyio | Not specified |
C | Not specified | everyfile |
D | everyio | everyfile |
E | everyio | everyfile |
F | Not specified | Not specified |
Specifies the identifier of the unit or the name of the host for which statistical information is to be output. If this option is omitted, statistical information for the entire HiRDB system is output to the statistics log file.
Specifies the servers for which statistical information is to be output. When this option is omitted, the statistical information to be output depends on the server type. For details about the statistical information that is output, see Table 2-36 Relationship between statistical information output and the -s option specification: HiRDB single server configuration or Table 2-37 Relationship between statistical information output and the -s option specification: HiRDB parallel server configuration.
Specifies that information about thread-to-thread lock-release wait time is to be collected as statistical information. The thread-to-thread lock-release wait time is the time required to acquire a thread-to-thread lock since a thread-to-thread lock request was placed in wait status.
You can collect information about thread-to-thread lock-release wait time as part of the following statistical information:
To collect information about thread-to-thread lock-release wait time, the system activity statistical information must also be collected. Therefore, if you are not collecting the system activity statistical information, collect it.
To collect more accurate information, execute the pdstjsync command immediately before or after you execute this command. For details, see 2.112 pdstjsync (Apply statistics log buffer to statistics log file).
To check the collection status of statistical information on the thread-to-thread lock-release wait time, execute the pdls -d stj command. For details, see 2.73 pdls [-d stj] (Display whether statistical data output type by unit and server is specified).
The following describes the conditions under which the statistical information on the thread-to-thread lock-release wait time can be collected for each statistical information type.
System activity statistical information (sys)
The following figure shows the period for which the system activity statistical information and the statistical information on the thread-to-thread lock-release wait time can be collected.
Figure 2-16 Period for which the statistical information on the thread-to-thread lock-release wait time can be collected (sys)
The period for which statistical information on the thread-to-thread lock-release wait time can be collected is from t3 to t4.
UAP statistical information (uap)
The following figure shows the period for which the UAP statistical information and the statistical information on the thread-to-thread lock-release wait time can be collected.
Figure 2-17 Period for which the statistical information on the thread-to-thread lock-release wait time can be collected (uap)
You can collect statistical information on the thread-to-thread lock-release wait time for UAPs that are started between the time the pdstbegin -w command is executed and the time the pdstend -w command is executed.
Global buffer pool statistical information (buf)
The following figure shows the period for which the global buffer pool statistical information and the statistical information on the thread-to-thread lock-release wait time can be collected.
Figure 2-18 Period for which the statistical information on the thread-to-thread lock-release wait time can be collected (buf)
After the pdstbegin -w command is executed, the period for which accurate statistical information on the thread-to-thread lock-release wait time can be collected is from synchronization point s2 that falls immediately after execution of the pdstbegin -w command to synchronization point s4 that falls immediately after execution of the pdstend -w command. In this example, the pdstend -w command is executed during transaction D, but the statistical information on the thread-to-thread lock-release wait time is collected until the transaction is completed.
If the unit contains a server that is not collecting statistical information on the thread-to-thread lock-release wait time, a value displayed as the result of editing the system activity statistical information for each unit will be smaller than the actual value. Therefore, if you limit the servers that will collect the system activity statistical information, evaluate the statistical information from the editing results for each server.
Table 2-36 Relationship between statistical information output and the -s option specification: HiRDB single server configuration
Types of statistical information | Servers specified with -s option (single server) |
---|---|
sys | Y |
uap | Y |
sql | Y |
sqh | Y |
buf | Y |
fil | Y |
dfw | Y |
idx | Y |
sop | Y |
dop | Y |
pcd | Y |
obj | -- |
dio | Y |
Table 2-37 Relationship between statistical information output and the -s option specification: HiRDB parallel server configuration
Types of statistical information | Servers specified with -s option | ||
---|---|---|---|
FES | DS | BES | |
sys | Y | Y | Y |
uap | Y | -- | -- |
sql | Y | -- | -- |
sqh | Y | -- | -- |
buf | -- | Y | Y |
fil | -- | Y | Y |
dfw | -- | Y | Y |
idx | -- | Y | Y |
sop | Y | -- | -- |
dop | Y | -- | -- |
pcd | Y | Y | Y |
obj | -- | Y | Y |
dio | N | Y | Y |
Figure 2-19 Relationship between output of statistical information and entry of the pdstbegin and pdstend commands
Server type | Server status | Target |
---|---|---|
Host BES | Executing | Y |
Standby | N | |
Guest BES | Executing | Y |
Acceptable status | N |