Analyzing Scheduler Job Management for Optimal Performance
Article ID: 53983
Updated On:
Products
Issue/Introduction
Scheduler has many features and options that may affect performance. A list of assessment techniques is provided to review the current usage of Scheduler.
Environment
Release: 11.0
Component: Scheduler Job Management MVS
Resolution
Assess the current statistics stored in the Scheduler database for usage of features and options. Below is the list of factors that can affect performance and ways to improve performance.
Dataset closes
Statistic ENF DSCLOSE shows the number of dataset close events received by Scheduler. When the DSCLOSE event is active, this number is usually extremely high. Statistic ENF DSCLOSE USE shows the number of those dataset close events that were actually used by Scheduler. This number is typically very small. Compare the number of DSCLOSE events used to the number actually received to show the incredible overhead inherent with dataset close events.
Review the criteria in use at the site to determine if any DSN or GDG keywords are in use. If not, then turn off the DSCLOSE event (either in ENF or with the Scheduler CAS9 ENF DSCLOSE=NO install option). If you find DSN or GDG keywords, investigate alternative methods for satisfying those requirements. For example, the POST DSN command can be used to satisfy a dataset requirement even with the DSCLOSE event turned off. Virtual resources can also be used instead of dataset requirements.
If the site requires dataset close processing, then contact ENF support for suggestions on using the SCREEN command to reduce the number of events to be processed by Scheduler.
JOBPRINT and JOBPURGE Events
Scheduler can use JOBPRINT and JOBPURGE events to automatically start and complete station 60 jobs. Most sites no longer use station 60 jobs, so the events are no longer necessary.
Statistics ENF JOBPRINT, ENF JOBPRINT USE, ENF JOBPURGE, and ENF JOBPURGE USE may be used to determine if Scheduler is using the events. JOBPURGE events, in addition to processing station 60 jobs, also mark completed station 40 jobs as "purged."
JOBPRINT and JOBPURGE events may be turned off in ENF or in Scheduler using the CAS9 ENF JOBPRINT=NO and JOBPURGE=NO options.
RDRQ Jobs
RDRQ (LIBTYPE=RDRQ) jobs are submitted by something other than Scheduler with a TYPRUN=HOLD on the job card. Scheduler must then find the job in JES and release it. The process of searching JES and releasing the job has very high overhead, especially when the job is not found and Scheduler must continue searching until the job is found.
Statistic MNTR RDRQ SEARCH reports how many times Scheduler looked for a RDRQ job. Statistic MNTR RDRQ NTFND reports how often the job was not found. The ratio of NTFND to SEARCH should be very low. A high ratio indicates excessive overhead.
Broadcom recommends an early start time or other tools be used to prevent Scheduler from looking for RDRQ jobs until after the job is likely to be in JES.
JES Performance
Scheduler uses an IBM service called ENDREQ at the end of each job submission to obtain the job's JES number. When JES is not tuned as well as it could be, the ENDREQ service can take a very long time for each job (sometimes as much as 8 seconds per job).
Statistic SUB ENDREQ TIME reports the total wall-clock time used by the ENDREQ service. Divide this number by the total number of jobs submitted (statistic SUB JOBS SUBBED) and multiple by .004096 to get the average ENDREQ time per job.
System Performance
Scheduler, ENF and DATACOM must be defined as non-swappable. Additionally, they all need to be in the same system performance group.
Installation Options
Option CMDQUE should be set to (NO,YES) to have batch commands (PGM=CAJUCMD0) use CMDQUE=NO by default (making sure each command completes before the next is evaluated), and TSO sessions use CMDQUE=YES by default (releasing the keyboard as soon as possible). Both batch jobs and TSO sessions can override the default.
Option DSNINFM should be set to NO. DSNINFM is only used with NJE=YES to send dataset close information to other JES nodes. The overhead of dataset close processing was discussed above.
Option PREDDSN should be set to NO. Using the PRED keyword with DSN criteria requires a great deal of processing on top of the processing required for dataset close events.
Option RETRY should be set to 60 seconds or higher. RETRY determines how often Scheduler queries JES for RDRQ jobs.
ENF Options
The ENF options should be set to exclude events for TSO sessions but to allow events for batch jobs and started tasks. In older releases of Scheduler, only batch jobs were tracked and it is possible for a site to have obsolete SELECT statements in ENF.
The statements should read:
SELECT(JOBINIT,JOBNUM,NE,TS*)
SELECT(STEPTERM,JOBNUM,NE,TS*)
SELECT(JOBTERM,JOBNUM,NE,TS*)
SELECT(JOBFAIL,JOBNUM,NE,TS*)
SELECT(JOBPRINT,JOBNUM,NE,TS*)
SELECT(JOBPURGE,JOBNUM,NE,TS*)
Additional Information
Feedback
