OpenVMS System Manager's Manual

Document revision date: 19 July 1999

OpenVMS System Manager's Manual

Contents

Index

25.5 Monitoring Transaction Performance

Changes to your system, such as increase in work load, can affect transaction performance. Once a month, monitor transactions on the node to make sure that transaction performance has not deteriorated. In an OpenVMS Cluster environment, monitor transaction performance on all the nodes in the cluster.

How to Perform This Task

Monitor transactions, using the MONITOR TRANSACTION command:

MONITOR TRANSACTION/SUMMARY[=file-spec]/ENDING=time/NODE=(nodename,...)

where:

file-spec is the file specification of the summary file. Information about transactions is summarized and recorded in the summary file. If you omit the file specification, the information is recorded in MONITOR.SUM in your default directory.

time is the time that the monitoring session ends.

nodename is the name of a node. In an OpenVMS Cluster, list all the nodes in the cluster.

For the best results, monitor transactions for a day at a time.
You can monitor transactions in batch mode by including the MONITOR TRANSACTION command in a command procedure.
For a full description of the MONITOR TRANSACTION command, refer to the OpenVMS System Management Utilities Reference Manual.

Examine the summary file.
The summary file contains values for a number of different data items. Note the following values for each node:
- Average end rate. This is the average number of transactions completed per second.
- Average completion rates. These are the average numbers of transactions completed in the following times:
  - Less than 1 second
  - Between 1 and 2 seconds
  - Between 2 and 3 seconds
  - Between 3 and 4 seconds
  - Between 4 and 5 seconds
  - More than 5 seconds
Keep a note of these values.
Compare the results from this monitoring session with the results from previous sessions.
For the same work load, the rate and duration of transactions should remain about the same. Indications of performance deterioration are:
- The average end rate decreases
- The average duration increases
  To find out whether the average duration of transactions has increased, compare the average completion rates. If a greater proportion of the transactions takes longer to complete, the average duration of transactions has increased.
Note any trends over a number of monitoring sessions. Variations from one monitoring session to the next are probably due to variations in work load.
If you suspect that transaction performance has deteriorated on any node, check whether its transaction log is too small (see Section 25.6).
If the transaction log is big enough, but transaction performance still deteriorates, tuning the system might be necessary. Refer to the OpenVMS Performance Management for information about tuning your system.

Example

This example shows how to monitor transaction performance on an OpenVMS Cluster that consists of two nodes whose SCSNODE names are BLUE and RED.

Monitor transactions on nodes BLUE and RED for one day:

$ MONITOR TRANSACTION/SUMMARY=DISK$LOG1:[LOGFILES]TRANSACTIONS.SUM - _$ /ENDING="+1-"/NODE=(BLUE,RED)

Examine the summary file:

DISTRIBUTED TRANSACTION STATISTICS on node BLUE From: 16-OCT-1998 14:23:51 SUMMARY To: 17-OCT-1998 14:23:51 CUR AVE MIN MAX Start Rate 49.02 43.21 31.30 49.02 Prepare Rate 48.70 43.23 30.67 48.70 One Phase Commit Rate 0.00 0.00 0.00 0.00 Total Commit Rate 48.70 43.19 31.30 48.70 Abort Rate 0.00 0.00 0.00 0.00 End Rate 48.70 43.19 31.30 48.70 Remote Start Rate 0.00 0.00 0.00 0.00 Remote Add Rate 0.00 0.00 0.00 0.00 Completion Rate 0-1 21.42 13.57 0.63 21.42 by Duration 1-2 25.97 29.15 24.59 33.87 in Seconds 2-3 1.29 0.47 0.00 4.47 3-4 0.00 0.00 0.00 0.00 4-5 0.00 0.00 0.00 0.00 5+ 0.00 0.00 0.00 0.00 SUMMARIZING DISTRIBUTED TRANSACTION STATISTICS on node RED From: 16-OCT-1998 14:23:52 SUMMARY To: 17-OCT-1998 14:23:52 . . .

Make a note of the following values:

Average end rate.
For node BLUE, the average end rate is 43.19 transactions per second.
Average completion rates.
For node BLUE, the average completion rates are as follows:
- 13.57 transactions completed in 0 to 1 seconds
- 29.15 transactions completed in 1 to 2 seconds
- 0.47 transactions completed in 2 to 3 seconds

Compare the results from this monitoring session to those of previous sessions:

Session End Rate Completion Rates

0--1 Secs 1--2 Secs 2--3 Secs

June 42.13 12.98 28.13 1.02

July 38.16 10.35 25.80 2.01

August 43.19 13.57 29.15 0.47

Session	End Rate	Completion Rates
		0--1 Secs	1--2 Secs	2--3 Secs
June	42.13	12.98	28.13	1.02
July	38.16	10.35	25.80	2.01
August	43.19	13.57	29.15	0.47

The results for node BLUE show no signs of deteriorating performance.

25.6 Checking Whether a Transaction Log Is Too Small

If transaction performance has deteriorated on a node, check whether its transaction log is too small.

Section 25.5 describes how to find out whether transaction performance has deteriorated.

How to Perform This Task

Log in to the node that the transaction log belongs to.
Check how many times the transaction log has stalled, using LMCP's SHOW LOG/CURRENT command:
$ RUN SYS$SYSTEM:LMCP LMCP> SHOW LOG/CURRENT
Note the number of checkpoints and stalls displayed by this command.
Wait for five minutes, then repeat the SHOW LOG/CURRENT command. Note the number of checkpoints and stalls again.
Compare the information from the SHOW LOG/CURRENT commands:
If the number of checkpoints has not changed, wait until the system is busier, then try this task again.
If the number of checkpoints has increased, and the number of stalls has increased by more than one, the transaction log is too small.
If the transaction log is too small, increase its size. For information about how to change the size of a transaction log, see Section 25.7.

Example

This example shows how to check whether node BLUE's transaction log is too small.

$ RUN SYS$SYSTEM:LMCP LMCP> SHOW LOG/CURRENT Checkpoint starts/ends 2464/2464 Stall starts/ends 21/21 Log status: no checkpoint in progress, no stall in progress.

The number of checkpoints is 2464, and the transaction log has stalled 21 times.

Wait for five minutes, then repeat the SHOW LOG/CURRENT command:

LMCP> SHOW LOG/CURRENT Checkpoint starts/ends 2514/2514 Stall starts/ends 28/28 Log status: no checkpoint in progress, no stall in progress.

The number of checkpoints has increased since the previous reading, and the transaction log has now stalled 28 times, an increase of 7. This means that the transaction log is too small.

25.7 Changing the Size of a Transaction Log

To determine if changing the size of a transaction log is necessary, see Section 25.6.

How to Perform This Task

Caution

Follow all the steps carefully. Taking shortcuts can lead to data corruption.

Log in to the node that the transaction log belongs to.
Find out which directory the transaction log is in, using LMCP's SHOW LOG command:
SHOW LOG SYSTEM$node.LM$JOURNAL
where node is the name of the node that the transaction log belongs to.

Rename the transaction log:

RENAME dirspecSYSTEM$node.LM$JOURNAL dirspecSYSTEM$node.LM$OLD

where:

dirspec is the full specification of the directory containing the transaction log.

node is the name of the node that the transaction log belongs to.

Can you stop all the software that uses DECdtm services without shutting down any nodes?

Yes

Close the transaction log as follows:

Step

Action

Stop all the software that uses DECdtm services.

Close the transaction log using LMCP's CLOSE LOG command:

$ RUN SYS$SYSTEM:LMCP

LMCP> CLOSE LOG

The CLOSE LOG command closes the transaction log and stops the DECdtm TP_SERVER process. The command fails if any software is using DECdtm services.

Did the CLOSE LOG command succeed?

Yes	Restart the TP_SERVER process: $ @SYS$STARTUP:DECDTM$STARTUP.COM
No	Wait for 30 seconds, then repeat steps 4b and 4c.

Close the transaction log by rebooting the node. Log in to the node when it has rebooted.

Change the size of the transaction log, using LMCP's CONVERT LOG command:

CONVERT LOG/SIZE=size dirspecSYSTEM$node.LM$OLD dirspecSYSTEM$node.LM$JOURNAL

where:

size is the new size of the transaction log in blocks.

dirspec is the full specification of the directory containing the transaction log.

node is the name of the node that the transaction log belongs to.

If you stopped the software that uses DECdtm services in step 4, restart the software.

Delete the old transaction log:

DELETE dirspecSYSTEM$node.LM$OLD;

where:

dirspec is the full specification of the directory containing the old transaction log.

node is the name of the node that the transaction log belongs to.

Example

This example shows how to change the size of node RED's transaction log to 6000 blocks. Node RED is in an OpenVMS Cluster, and its transaction log is in DISK$LOG2:[LOGFILES].

$ RUN SYS$SYSTEM:LMCP LMCP> SHOW LOG SYSTEM$RED.LM$JOURNAL Directory of DISK$LOG2:[LOGFILES] SYSTEM$RED.LM$JOURNAL;1 Total of 1 file. LMCP> EXIT $ RENAME DISK$LOG2:[LOGFILES]SYSTEM$RED.LM$JOURNAL - _$ DISK$LOG2:[LOGFILES]SYSTEM$RED.LM$OLD

Stop all software that uses DECdtm services. Then close the transaction log:

$ RUN SYS$SYSTEM:LMCP LMCP> CLOSE LOG Transaction log closed, TP_SERVER process stopped LMCP> EXIT

Restart the TP_SERVER process:

$ @ SYS$STARTUP:DECDTM$STARTUP.COM

Change the size of the transaction log:

$ RUN SYS$SYSTEM:LMCP LMCP> CONVERT LOG/SIZE=6000 DISK$LOG2:[LOGFILES]SYSTEM$RED.LM$OLD - _LMCP> DISK$LOG2:[LOGFILES]SYSTEM$RED.LM$JOURNAL Log file DISK$LOG2:[LOGFILES]SYSTEM$RED.LM$JOURNAL;1 created. Log file DISK$LOG2:[LOGFILES]SYSTEM$RED.LM$OLD converted. LMCP> EXIT

Restart the software that uses DECdtm services.

Delete the old transaction log:

$ DELETE DISK$LOG2:[LOGFILES]SYSTEM$RED.LM$OLD;

25.8 Moving a Transaction Log

You may want to move a transaction log if:

You want to place the transaction log on a faster disk
You want to redistribute the work load on your disks

How to Perform This Task

Caution

Follow all the steps carefully. Taking shortcuts can lead to data corruption.

Decide the location that you want to move the transaction log to, using the guidelines in Section 25.2.2. Remember that the disk must have enough contiguous space to hold the transaction log.
Log in to the node that the transaction log belongs to.
If you are in an OpenVMS Cluster, make sure that the disk you want to move the transaction log to is mounted clusterwide.
Decide which directory you want to move the transaction log to. You may want to create a new directory for the transaction log.
Find out which directory the transaction log is in, using LMCP's SHOW LOG command:
SHOW LOG SYSTEM$node.LM$JOURNAL
where node is the name of the node that the transaction log belongs to.

Rename the transaction log:

RENAME dirspecSYSTEM$node.LM$JOURNAL dirspecSYSTEM$node.LM$OLD

where:

dirspec is the full specification of the directory containing the transaction log.

node is the name of the node that the transaction log belongs to.

Can you stop all the software that uses DECdtm services without shutting down any nodes?

Yes

Close the transaction log as follows:

Step

Action

Stop all the software that uses DECdtm services.

Close the transaction log using LMCP's CLOSE LOG command:

$ RUN SYS$SYSTEM:LMCP

LMCP> CLOSE LOG

The CLOSE LOG command closes the transaction log and stops the DECdtm TP_SERVER process. The command fails if any software is using DECdtm services.

Did the CLOSE LOG command succeed?

Yes	Restart the TP_SERVER process: $ @SYS$STARTUP:DECDTM$STARTUP.COM
No	Wait for 30 seconds, then repeat steps 7b and 7c.

Close the transaction log by rebooting the node. Log in to the node when it has rebooted.

Make sure that SYS$JOURNAL points to the directory that you want to move the log to. If SYS$JOURNAL does not point to this directory, redefine SYS$JOURNAL:
DEFINE/SYSTEM/EXECUTIVE_MODE SYS$JOURNAL dirspec[,...]
where dirspec is the full specification of a directory containing one or more transaction logs. List all the directories that will contain transaction logs after you have moved the transaction log. You can list the directories in any order.
In an OpenVMS Cluster, use SYSMAN to redefine SYS$JOURNAL clusterwide.
If you redefined SYS$JOURNAL in step 8, edit the SYS$MANAGER:SYLOGICALS.COM command procedure to update the definition of SYS$JOURNAL.
If you created node-specific versions of SYLOGICALS.COM, edit all the versions.

Move the transaction log, using LMCP's CONVERT LOG command:

CONVERT LOG old-dirspecSYSTEM$node.LM$OLD new-dirspecSYSTEM$node.LM$JOURNAL

where:

old-dirspec is the full specification of the directory that currently contains the transaction log.

node is the name of the node that the transaction log belongs to.

new-dirspec is the full specification of the directory that you are moving the transaction log to.

If you stopped the software that uses DECdtm services in step 7, restart the software.

Delete the old transaction log:

DELETE dirspecSYSTEM$node.LM$OLD;

where:

dirspec is the full specification of the directory containing the old transaction log.

node is the name of the node that the transaction log belongs to.

Example

This example shows how to move BLUE's transaction log. BLUE is in an OpenVMS Cluster. The cluster members and the locations of their transaction logs are as follows:

Node Directory Containing Log

BLUE DISK$LOG1:[LOGFILES]

RED DISK$LOG2:[LOGFILES]

Node	Directory Containing Log
BLUE	DISK$LOG1:[LOGFILES]
RED	DISK$LOG2:[LOGFILES]

Neither node has a node--specific version of SYLOGICALS.COM.

Decide where you want to move BLUE's transaction log to. In this example, assume that you want to move it to DISK$LOG3:[LOGFILES].

$ MOUNT/CLUSTER/SYSTEM DUA3: LOG3 $ CREATE/DIRECTORY DISK$LOG3:[LOGFILES]

Find out which directory BLUE's transaction log is in, then rename the transaction log:

$ RUN SYS$SYSTEM:LMCP LMCP> SHOW LOG SYSTEM$BLUE.LM$JOURNAL Directory of DISK$LOG1:[LOGFILES] SYSTEM$BLUE.LM$JOURNAL;1 Total of 1 file. LMCP> EXIT $ RENAME DISK$LOG1:[LOGFILES]SYSTEM$BLUE.LM$JOURNAL - _$ DISK$LOG1:[LOGFILES]SYSTEM$BLUE.LM$OLD

Stop all software that uses DECdtm services. Then close the transaction log:

$ RUN SYS$SYSTEM:LMCP LMCP> CLOSE LOG Transaction log closed, TP_SERVER process stopped LMCP> EXIT

Restart the TP_SERVER process:

$ @SYS$STARTUP:DECDTM$STARTUP.COM

Redefine SYS$JOURNAL:

$ RUN SYS$SYSTEM:SYSMAN SYSMAN> SET ENVIRONMENT/CLUSTER SYSMAN> DO DEFINE/SYSTEM/EXECUTIVE_MODE SYS$JOURNAL - _SYSMAN> DISK$LOG2:[LOGFILES], DISK$LOG3:[LOGFILES] SYSMAN> EXIT

Edit the SYS$MANAGER:SYLOGICALS.COM command procedure to update the SYS$JOURNAL definition. Then move the transaction log:

$ RUN SYS$SYSTEM:LMCP LMCP> CONVERT LOG DISK$LOG1:[LOGFILES]SYSTEM$BLUE.LM$OLD - _LMCP> DISK$LOG3:[LOGFILES]SYSTEM$BLUE.LM$JOURNAL Log file DISK$LOG3:[LOGFILES]SYSTEM$BLUE.LM$JOURNAL;1 created. Log file DISK$LOG1:[LOGFILES]SYSTEM$BLUE.LM$OLD converted. LMCP> EXIT

Restart the software that uses DECdtm services. Then delete the old transaction log:

$ DELETE DISK$LOG1:[LOGFILES]SYSTEM$BLUE.LM$OLD;

Contents

Index

privacy and legal statement

6017PRO_100.HTML

file-spec	is the file specification of the summary file. Information about transactions is summarized and recorded in the summary file. If you omit the file specification, the information is recorded in MONITOR.SUM in your default directory.
time	is the time that the monitoring session ends.
nodename	is the name of a node. In an OpenVMS Cluster, list all the nodes in the cluster.

dirspec	is the full specification of the directory containing the transaction log.
node	is the name of the node that the transaction log belongs to.

size	is the new size of the transaction log in blocks.
dirspec	is the full specification of the directory containing the transaction log.
node	is the name of the node that the transaction log belongs to.

old-dirspec	is the full specification of the directory that currently contains the transaction log.
node	is the name of the node that the transaction log belongs to.
new-dirspec	is the full specification of the directory that you are moving the transaction log to.