Easytrieve Performance Best Practices

Measure Before You Tune

Every tuning story starts with baseline numbers: job elapsed time, CPU seconds, EXCP counts, and sort messages. If EXCP dominates and CPU is low, you are likely I/O-bound—buffering, file design, and reducing passes over data help first. If CPU is high with modest I/O, look at expressions in tight loops, table searches, and unnecessary conversions. If sort steps or WORK file allocations spike, focus on keys, duplicate handling, and whether presorted input eliminates an internal sort. Changing twenty options without classification wastes change-control cycles and can regress a stable job. Capture before and after in a tuning log the operations team can read six months later.

Bottleneck Quick Reference

I/O Best Practices

Sequential FILE definitions accept environmental options including BUFNO—the number of buffers for QSAM reads and writes. A higher BUFNO can reduce physical reads when the job scans the same large file once sequentially. It is not free: each buffer consumes memory below the 16 MB line in classic AMODE31 load modules, and dozens of files with BUFNO 255 can exhaust storage. Best practice: identify the one or two largest input files, test BUFNO incrementally (for example 5, 10, 20), and stop when elapsed time flattens. VSAM files use different mechanisms; blindly copying BUFNO from a colleague's sequential job helps nothing on KSDS. Avoid reading the same file multiple times in one job when a single pass with WORK file or SORT can produce both detail and summary outputs.

1
2
3
4
FILE INPUT-FILE
  FILE INPUT-FILE
  FIELDS
  BUFNO 10

Place BUFNO on the FILE environmental options where your release documents it—verify against your Broadcom manual because syntax varies slightly by version. Combine with sensible BLKSIZE and dataset organization chosen at allocation time, not only in Easytrieve source.

VIRTUAL Files and VFM

VIRTUAL files store intermediate data in Virtual File Manager temporary space. They simplify program logic when you need scratch ordering or multiple passes without permanent datasets. Under load, VFM contention slows every job on the LPAR. Best practices: allocate adequate VFMSPAC in startup or job PARM, delete virtual files when no longer needed, and prefer direct sequential WORK datasets for very large scratch when operations standards allow. Monitor temp space during peak batch windows before adding new VIRTUAL-heavy programs to the schedule.

SORT and WORK File Discipline

SORT activities and SORT statements inside jobs consume CPU and disk for keys and merge passes. Presorting extract files in a dedicated DFSORT step often beats repeated internal sorts across similar jobs. When you must sort inside Easytrieve, specify only necessary key fields—extra key bytes increase work. DUP handling choices affect whether you process duplicate keys twice or collapse them early. WORK file space in REPORT activities can explode when control breaks multiply; design break levels deliberately. See sort optimization and multi-key sorting tutorials for field-level detail.

CPU and Logic Best Practices

• Remove production DISPLAY except sampled counters (every N records) when audit requires.
• Move invariant calculations outside loops—compute rate once per job, not per record.
• Use table LOOKUP with sorted keys and binary search patterns for large reference tables.
• Avoid repeated string trims and conversions; define fields with correct types upfront.
• Split rarely used exception paths into PERFORM modules so the common path stays lean.

DEBUG and FLOW options belong in development compiles. Shipping DEBUG FLOW to production multiplies overhead and fills listings operators ignore until disk fills. Use controlled test jobs with limited input when tracing logic.

Compiler and Runtime Options

Compiler optimization options (site-specific names vary by release) trade compile time and listing detail for tighter generated code. Standardize production compile procedures with optimization enabled after regression test. Cross-reference and map listings help performance work indirectly by finding dead fields and accidental duplicate definitions that bloat work storage. Keep listing options that operations require for audit, but do not enable every diagnostic in nightly production rebuilds without reason.

Report and Online Considerations

Batch report generators spend much time in REPORT activities building lines and control breaks. Minimize arithmetic in BEFORE-LINE when AFTER-BREAK can accumulate once per group. Online SCREEN programs pay terminal I/O and conversational overhead—performance tuning targets round trips and file GET counts in BEFORE-SCREEN, not BUFNO on batch files. Match tuning technique to program type; batch habits pasted onto CICS-backed screens misallocate effort.

Production Change Checklist

1. Record baseline elapsed time, CPU, EXCP, and sort disk from SMF or job output.
2. Change one category at a time (I/O, sort, logic) when possible.
3. Regression test functional output—not just runtime—totals must still tie.
4. Document options changed in program header comment and change ticket.
5. Roll back quickly if storage abends appear after BUFNO or WORK increases.

Explain It Like I'm Five

Performance tuning is making sure the delivery truck takes the fast road without breaking the bridge. First you watch whether the truck is slow because of traffic (disk reads), heavy boxes (sorting piles), or the driver checking every single box out loud (DISPLAY). You fix the biggest problem first—maybe take a highway with more lanes (buffers)—but you do not add a hundred lanes if the bridge collapses (memory abend). You measure, change one thing, and see if the truck arrives earlier before changing everything else.

Exercises

1. From a sample job log, classify I/O-bound versus CPU-bound and justify your answer.
2. List three DISPLAY alternatives for debugging a million-record loop.
3. Explain when presorted input removes an internal SORT step.
4. Draft a BUFNO test plan for one large sequential file with three buffer values.
5. Name two report-phase habits that reduce WORK file growth.