Easytrieve File Buffering — BUFNO Tuning

What BUFNO Does

Each buffer holds one block of data read or written for a sequential file. Multiple buffers allow read-ahead: while your program processes records from buffer one, I/O subsystem fills buffer two. Write path similarly pipelines block output. Diminishing returns appear when buffers exceed what sequential scan can consume before file end—measure rather than assume 255 is optimal.

Three Places to Set BUFNO

Input Versus Output Precedence

Broadcom knowledge article 54905 documents different parse rules. Input files: JCL DCB BUFNO if present; else FILE BUFNO; else Options Table. Output files: JCL ignored; FILE BUFNO; else Options Table. Beginners set JCL BUFNO on output DD wondering why behavior unchanged— move BUFNO to FILE for output datasets.

1
2
3
4
5
6
7
8
* Source
FILE PAYIN  FB(150 1500) BUFNO 8
FILE PAYOUT FB(150 1500) BUFNO 12
 
* JCL — BUFNO on PAYIN DD overrides FILE 8 if coded
//PAYIN   DD DSN=PAYROLL.IN,DISP=SHR,DCB=(BUFNO=20,...)
//POUT    DD DSN=PAYROLL.OUT,DISP=(NEW,CATLG)
* POUT JCL BUFNO ignored — uses FILE BUFNO 12

FILE Statement Syntax

FILE Statement documents BUFNO 1 to 255 for MVS. Value establishes buffer count allocated for that file definition. Pair with realistic FB BLKSIZE—buffer size follows block structure; mismatched BLKSIZE in FILE versus JCL DCB causes confusion unrelated to BUFNO count.

Experimental Tuning Method

1. Baseline job with default BUFNO—record EXCP and elapsed.
2. Double BUFNO on largest input file only; re-run same input.
3. If EXCP drops materially and no storage abend, iterate upward.
4. Repeat for heavy output file using FILE BUFNO.
5. Stop when marginal EXCP reduction per buffer added is small.

AMODE31 and Below-the-Line Storage

When many files carry elevated BUFNO, total dynamic buffer storage grows. Options Table AMODE31 controls whether dynamic allocations sit above or below 16MB line. Article 16520 describes program that abended below-the-line after BUFNO 5 on all sequential files—linked AMODE 31 RMODE ANY with AMODE31=NO placing buffers below line. Coordinate with systems programming before mass BUFNO change on large multi-file programs.

VSAM and VFM Exceptions

Environmental options explicitly state VFM and VSAM files do not use BUFNO. VSAM buffering uses AMP parameters in JCL or cluster CI size tuning. VIRTUAL files use VFM paths—see performance overview VFM section. Applying BUFNO to VSAM FILE definition does not help sequential browse performance.

BUFNO Versus Design Changes

Buffering cannot eliminate extra pass when JOB reads file, SORT writes work, second JOB reads again—design merge may dominate. Presort SELECT, combined activities, or indexed lookup versus full scan are architectural wins BUFNO alone cannot deliver.

Common File Buffering Mistakes

• BUFNO on VSAM expecting same effect as QSAM.
• JCL BUFNO on output wondering why FILE value still applies incorrectly—output ignores JCL.
• Maximum BUFNO everywhere causing storage abend.
• Tuning BUFNO on tiny control file with ten records—no measurable gain.
• Ignoring BLKSIZE mismatch while tuning buffers.
• Not re-measuring after LPAR storage parameter changes.

Explain It Like I'm Five

BUFNO is how many armfuls of papers you carry from the file cabinet at once. Two armfuls means many trips; ten armfuls means fewer trips but your arms get full—too many armfuls for every drawer at once and you cannot hold them all. Input homework can also say in JCL how many armfuls; output homework only listens to the FILE line.

Exercises

1. Add BUFNO 10 to largest input FILE; document precedence with JCL DCB.
2. Explain why output ignores JCL BUFNO in one paragraph.
3. List three file types BUFNO does not help.
4. Design tuning experiment with baseline and one-variable BUFNO change.
5. Describe AMODE31 relevance when adding BUFNO to fifteen files.