1. Adabas Performance Topics
Jim Poole, Verizon
April, 2014

2. CONTENTS Forward Index Compression MTR for ADARES REGENERATE
RUN UTILITIES M.PM., NOT S.U.M.
USE ADARUN LFIOP
USE ADAREP NOCOUNT AS A DEFAULT
PE GROUPS, ALWAYS ADD UNUSED FIELDS
INCREASE BLOCKSIZE
DEFAULT BATCH MLF
KNOW WHEN TO CACHE, WHAT NOT TO
SHADOW DATABASES, 7X24
MAXIMIZE WORK PART I (ADARUN LP=)
USE ADARUN NPLOG/NWORK1
LIMIT USER VIEW SIZE
USE S/J OR DIRTY READ ZAP
ADARUNS TO MINIMIZE
DO NOT USE FIXED COMPRESSION
AVOID EXCESSIVE REORDERING
PADDING AND UNUSED FILE SPACE

3. 1. Forward Index Compression
SET WITH ADALOD or ADAORD
INDICATOR IN ADAREP
SAVE INDEX SPACE (40 – 50%)
SAVE ON DATABASE GROWTH
REDUCE IO FOR L3/L6 L9
IMPROVED BUFFER EFFICIENCY
FASTER UTILITIES
NO INCREASE CPU OVERHEAD

4. 1. Forward Index Compression
FIC->

5. 1. Forward Index Compression
FIC

6. 1. Forward Index Compression
FIC->

7. 2. MTR for ADARES REGENERATE
//DDKARTE
MTR=YES
DATABASE ADARUN:
NT= (to allow overlap)
NABS= (NT*16, else !ERROR-099! abend may occur)
NC= (>2*NT)
LFIOP= (for performance)
MULTI-THREADING PROCESSING STATISTIC
PLOG BLOCKS READ FROM INPUT
PLOG RECORDS SENT TO ADABAS
COMMANDS PROCESSED
TRANSACTIONS PROCESSED
NUMBER OF ADABAS CALLS
MAXIMUM CALLS IN PARALLEL
AVERAGE CALLS IN PARALLEL
AVERAGE RECORD BUFFER SIZE

8. 2. MTR for ADARES REGENERATE
MTR=NO 
MTR=YES 

9. 3. RUN UTILITIES M.PM., NOT S.U.M
S.U.M.: NO CACHE, ALL BELOW 16M LINE
CONSIDER UTIONLY=YES
EXAMPLE: ADARES REGENERATE :
-SUM MPM--
ELAPSED ELAPSED
TOTAL= M M 80% FASTER
AVER== M M

10. 4. USE ADARUN LFIOP ALLOWS ASYNCHRONOUS PROCESSING
LARGER LFIOP DECREASES BUFFER FLUSHES
BUFFER FLUSHES
LFIOP= 1M 2M 4M 6M 8M 10M 12M 20M 25M

11. 5. USE ADAREP NOCOUNT AS A DEFAULT
ADAREP BY DEFAULT READS EVERY USED AC BLOCK FOR RECORD COUNT
DB HAS 60 FILES, 100k RECORDS
ADAREP ‘COUNT’: 122K ASSO IO, 4.65 MINS.
ADAREP NOCOUNT: ASSO IO, .46 MINS.
5,483,439,070 Records
6,487,490 AC used blocks
$25K CPU/year for Adarep
ALTERNATE IDEA
FNDEF='01,FC,1,A,DE' FILE-COUNTER (NO NU)
HISTOGRAM for RECORD COUNT
ADAREP READS AC ON DISK
HISTOGRAM READS FROM LBP (BUFFER POOL)

12. 6. PE GROUPS, ALWAYS ADD UNUSED FIELDS
RESERVE FIELDS FOR FUTURE USE
ADACMP FNDEF='01,AO,5,U,NU' TERMN-ID
ADACMP FNDEF='01,BA,PE(191)' BRANCHES
ADACMP FNDEF='02,BB,14,U,NU' NEXT-NODE-ID
ADACMP FNDEF='02,BC,3,A,NU' NEXT-NODE-TYPE
ADACMP FNDEF='02,BD,1,A,NU' DEFAULT-IND
ADACMP FNDEF='02,BE,1,A,NU' NOT-USED-001
ADACMP FNDEF='02,BF,1,A,NU' NOT-USED-002
ADACMP FNDEF='02,BG,1,A,NU' NOT-USED-003
ADACMP FNDEF='02,BH,1,U,NU' NOT-USED-004
ADACMP FNDEF='02,BI,1,U,NU' NOT-USED-005
ADACMP FNDEF='02,BJ,1,U,NU' NOT-USED-006

13. 7. INCREASE BLOCKSIZE a. EXAMPLE: 8391 OR 8393
b. CAN INCREASE SPACE AVAILABLE PER CYLINDER
c. AVOID RESPONSE CODE 53
d. ALTERNATE TO SPANNED RECORDS
e. MAY IMPROVE PERFORMANCE ON READS
EXAMPLE: DEVICE = 8391
LBP = 2 X OLD.LBP
LFIOP = 2 X OLD.LBP
WORK: EITHER DOUBLE WORKSIZE or
LP = .42 X OLD.LP
LWKP2 = .42 X OLD.LWKP2
DUALPLD = 8391
DUALPLS = 60 X PLOG.CYLINDERS
CACHE PARMS

14. 8. DEFAULT BATCH MLF SET UP AS DEFAULT FOR ALL BATCH JOBS
ADARUN PREFETCH=YES
ADARUN PREFSBL=60000
ADARUN PREFTBL=240000
ADARUN PREFNREC=2
( SBL SIZES ARE OPTIMUM FOR PRE V8 NON-ACBX )

15. 14% CPU SAVINGS ($25k CPU COSTS PER YEAR)
8. DEFAULT BATCH MLF
RESULTS
14% CPU SAVINGS ($25k CPU COSTS PER YEAR)

16. QUICK CACHE NOTES WHEN TO CACHE ASSO/DATA?
1. SYSTEM MEMORY CONSTRAINTS FOR ADABAS TASK
(eg. Paging, availability of 64bit, etc)
2. LIMITS TO LBP SIZE DUE TO FREQUENT BUFFER FLUSHES.
(cpu time to scan for updated blocks);
(otherwise no limits to LBP size except for #1 above.)
3. ADVANTAGE OF AN ‘EXTENDABLE’ BUFFER POOL
(can resize based on demand)
4. IMPROVE PERFORMANCE ON SELECTED FILES OR JOBS.
5. ‘PRIORITIZE’ RETENTION OF CACHE BLOCKS (service levels)
6. PERFORMANCE ON DEMAND TO USE CBUFNO

17. QUICK CACHE NOTES FIRST. WHEN TO CACHE? WORK
W WORK II - ALWAYS (SEE SAG WIKI)
WORK III - ALWAYS (SEE SAG WIKI) WW
ASSO/DATA – GENERALLY DO NOT CACHE
“IN GENERAL X MEG FOR LBP IS BETTER THAN X MEG FOR CACHE”
GOAL OF CACHING
“The goal of Adabas Caching is to improve response time by reduced IO without incurring unacceptable CPU increases or unacceptable memory usage (conflicting with other non-Adabas users of the system).”

18. QUICK CACHE NOTES < 30% 30 – 60 % > 60 % CACHE EFFICIENCY
MORE CPU
SLOWER RESPONSE
MORE CPU
LESS CPU
FASTER RESPONSE
FASTER RESPONSE

19. Goal: Response time (not IO)
QUICK CACHE NOTES
WHAT'S BETTER: ASSO, DATA, OR BOTH?
1. Best to cache DATA or possibly BOTH
2. As the amount of ASSO IO increases, then there is an increasing tendency to cache BOTH
3. Rarely cache only ASSO.
Goal: Response time (not IO)
(TESTED UP TO WHERE 50% OF ALL IO FROM ASSO)

20. 10. SHADOW DATABASES, 7X24 AVOID OUTAGES
1. CONVERT FILES TO 4B ISN OR REVERSE. (EXANDED FILE REDO)
2. CHANGE BLOCKSIZE OF DATABASE
3. REORDER TO FORWARD INDEX COMPRESSION
4. COMPLETE REORDER (empty Index blocks, extents, cluster, resequ.)
5. HIGH VOLUME TESTING. DBA TESTING
6. DATABASE INTEGRITY CHECKS. INTEGRITY FIX: ULD/LOD
7. CONVERT SPANNED FILES TO UNSPANNED
8. SOME MAJOR APPLICATION CHANGES
9. MERGE/SPLIT DATABASES
10. DISASTER RECOVERY SITE

21. 10. SHADOW DATABASES, 7X24 ADARES REGEN PRIMARY DB 99 SHADOW GAIN:
PLOG
ADARES REGEN
PRIMARY
DB 99
SHADOW
GAIN:
MAXIMUM OUTAGE
TIME: 5 – 15 MINS.
COST:
DUPE DASD FOR
ASSO/DATA/WORK
SETUP TIME
MONITOR TIME
CPU: < 2%
ALTERNATIVE:
EVENT REPLICATOR
COST:
SETUP TIME
CPU
LICENSE COST

22. 11. MAXIMIZE WORK PART 1
FOR MORE DETAILS ON SIZING, EXPLANATIONS, SEE:
SAG WIKI Article
"WORK Dataset. Sizing, Performance, Issues”
WORK
DEVICE=8393 (half track)
ADARUN LP=89,000
FOR EXAMPLE
5,000 CYLS

23. 11. MAXIMIZE WORK PART 1 NUC PGM FILE SMALL LP INCREASES
BUFFER FLUSHES
SMALL LP INCREASES
BUFFER FLUSHES
FOR EXAMPLE
LBP=100M
LFIOP=25M
LP=???
NUC
READ FILE:
L5,A1,ET
(LOOP 10X)
PGM
BLOCKS WRITTEN
20K RECORDS
999 DS BLKS
BLKSIZE=
10796
FILE
LP= K 26K 41k

24. ADARUN NPLOG/NWORK1 (ADARUN)
ADARUN NPLOGBUFFERS = 1 - 2,147,483,687
ADARUN NWORK1BUFFERS = 1 - 2,147,483,687
1. DEFAULT = 1 (1 IO PER BLOCK)
2. REQUIRES LFIOP /= 0
3. IF ADARUN V64BIT=YES Then ABOVE 2G BAR
4. MEMORY SPACE = buffer.count X blocksize

25. 12. USE ADARUN NPLOG/NWORK1
BATCH JOB DURATION
(HIGH UPDATE)
BUF=1
BUF=10
BUF=40
TEST:

26. ! AVOID GENERALIZED VIEWS !
13. LIMIT USER VIEW SIZE
! PROGRAMS SHOULD SELECT ONLY THE FIELDS NEEDED !
! AVOID GENERALIZED VIEWS !
GENERAL : READS 214 FIELDS OFF FILE
B. SPECIFIC: READS 20 FIELDS OFF FILE
DO 100,000 READS WITH BOTH METHODS.
EXAMPLE

27. 13. LIMIT USER VIEW SIZE RESULTS: PERCENT SPECIFIC GENERAL CHANGE
EXAMPLE
PERCENT
SPECIFIC GENERAL CHANGE
RECORD BUFFER 115b b
ELAPSED TIME MIN MIN %
CPU OF USER ? ? %
CPU OF ADABAS 89 SEC SEC %
ADABAS SQL: AVOID SELECT *
USE NATURAL LDA’S FOR ONLY THE NEEDED FIELDS

28. 14. USE S/J OR DIRTY READ ZAP
USER OBSERVATIONS
WITHOUT THE ZAP
30%
FOVW:
FROM 0 TO
1000/WK
12%
FORMAT OVERWRITES
LFP MAXED
12% 3%
20% ?
10-15%

29. 14. USE S/J OR DIRTY READ ZAP
DB X CPU TIME PER CMD
37 DAYS BEFORE/AFTER
SORTED LOW TO HIGH
NO ZAP =>
WITH ZAP

30. 14. USE S/J OR DIRTY READ ZAP
DIRTY READ CHECK, RC 175
“ UNDER V7.4 ON AN L3, THE VALUE IN THE INDEX IS
CHECKED AGAINST THE VALUE IN DATA STORAGE.
IF THEY DO NOT MATCH, A RC 175 IS RETURNED.”
L3,L6, some S1/4
Response 175
Explanation: An inconsistency was detected between
the index and Data Storage.
Action: Run the "check" utilities (especially ADAICK and
ADAVAL) against the file, and contact your Software AG
technical support representative.
PER MESSAGES AND CODES MANUAL (NEW TO 7.4):

31. 14. USE S/J OR DIRTY READ ZAP
MODULES: ADANC5/ADANC7/ADANC9/ADARUN
PROBLEM: PERFORMANCE ZAP: WHEN NUCID=0
SKIP THE S/J TEST THAT USES THE VALUE IN THE
COMPRESSED RECORD. INSTEAD SCAN THE THREADS
FOR AN ACTIVE UPDATE FOR THE SAME FNR/ISN.
KEYWORD: PERFORMANCE NON-CLUSTER SPECIAL

32. 15. ADARUNS TO MINIMIZE “SHRINK THEM UNTIL THEY COMPLAIN” PARM
RESPONSE CODE
USER OVERRIDE
MAX SETTING
NOTE TT 9
YES
MXTT
TRANSACTION TIME LIMIT
TNAE
MXTNA
NON-ACTIVITY TIME LIMIT CT
254 NO
INTER-REGION TIME LIMIT
NISNHQ 47
MAX RECORDS ON HOLD
NQCID 46
MAX CID'S PER USER

33. 16. DO NOT USE FIXED COMPRESSION
FIXED FIELDS :
SAVE ONE BYTE PER FIELD
CANNOT CHANGE LENGTH LATER
DOES NOT SAVE ON CPU

34. 16. DO NOT USE FIXED COMPRESSION
KEY: NU
FIXED
DfLT
Test runs

35. 17. AVOID EXCESSIVE REORDERING
IN GENERAL:
FILES WILL BEGIN TO REUSE SPACE
DOES NOT IMPROVE OVERALL CPU/IO’S

36. 18. PADDING AND UNUSED FILE SPACE
DO THIS SO I CAN SES
SET ASSO/DATA PAD TO 1%
LET USE CREATE FREE SPACE
NEW THOUGHTS, NOT FINALIZED
DO THIS SO I CAN SES
REDUCE FILE FREE SPACE
TAKE MANY FILE EXTENTS