ADABAS TIPS GUTS PERFORMANCE

1
ADABASTIPSGUTS PERFORMANCE
Jim Poole Fall, 2005
2
TOPICS A. IMPROVED LFP PROCESSING B. HIGH
UPDATES AND CHECKPOINTS C. QBLKSIZE AGAIN D.
DIRTY READS E. ADARVU F. AOS BATCHYES G. 64 BIT
CACHING H. CACHE GLOBAL vs FILE I. CACHE IS IT
DEAD? J. PPT K. PARM-ERROR-74 L. PARM-ERROR-72 M.
RSP.CD. 199, FIXING INTEGRITY ERRORS N. LARGER
BLOCKSIZES O. EXCESSIVE REORDERING P. FASTER PLOG
COPIES
3
D. DIRTY READS
THE OBSERVATIONS GOING TO V742
30
12 FORMAT OVERWRITES LFP MAXED
FOVW FROM 0 TO 1000/WK
12 3 20 ?
10-15
4
D. DIRTY READS
THE OBSERVATIONS GOING TO V742
DB X CPU TIME PER CMD 37 DAYS BEFORE/AFTER SORTED
LOW TO HIGH
AFTER gt
BEFORE
5
D. DIRTY READS
CONTENTS OF DISCUSSION.
1. NEW TO V7.4 DIRTY READ CHECK RC 175 2.
REVIEW OF BPM, THREAD LOGIC, CONCURRENCY 3. HOW
TO CAUSE AN INTEGRITY ERROR 4. V71 vs V74 FOR
INTEGRITY ERROR 5. COST OF DIRTY READ CHECKING
6. COST OF DECOMPRESS 1X OR 2X 7. COST OF LFP,
IFB 8. IFB PROBLEM WITH SUB,SUPER DE 9. COST OF
SUB,SUPER DE IN DIRTY READ CHECKING 10. IFB
PROBLEM WITH L3 SHARED GFID. 11. Sx. WHICH DOES
DIRTY READ CHECKING 12. Sx. WHICH DOES NOT DO
CHECKING 13. Sx. READ OF ISN LIST 14. Sx. LFP,IFB
COSTS 15. Sx. READ OF ISN LIST AND IFB 16. Sx.
LAB STUDYS,THINGS GONE BAD 17. Eg. SHOPS THAT
WILL REALLY SUCK 18. SUMMARY OF COSTS, ISSUES,
WHERE SAG CAN IMPROVE.
6
D. DIRTY READS
DIRTY READ CHECK, RC 175
PER SAG, BOSTON CONF. 2004
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
7
D. DIRTY READS
REVIEW
CONCURRENCY, BUFFER POOL MANAGER, THREAD LOGIC
BPM
L3
A1
8
D. DIRTY READS
REVIEW
FOR EXAMPLE
CONCURRENCY, BUFFER POOL MANAGER, THREAD LOGIC
USERA
FOR 100K TIMES A1ISN888,NMSMITH BT
USERB
FOR 500K TIMES L3COP2V,VBJONES WRITE ISN
NM RESET
9
D. DIRTY READS
Example
CAUSE AN INTEGRITY ERROR
(ACTIONS ON ISN 7, SAME FILE)
3.RI ISN 7
10
D. DIRTY READS
Example cont.
L3 BY NAME WRITE NAME
lt V7.4, eg V7.1
V7.4
ADAMS ADAMS FINK JOHN SMITH KIRK ZIRK
DIRTY READ ALLOWED
11
D. DIRTY READS
WHAT IS THE COST OF L3/L6 CHECKING DS?
?
RUN PGM 100K (ONLY 7 IOS TOTAL) CYCLE DB MEASURE
CPU,ETC V7.1 vs V7.4?
12
D. DIRTY READS
AA,10,A,DE AB,15,N ZZ,03,A,DE (Fld 262)
FDT
DOES IT DECOMPRESS 2X OR 1X?
13
D. DIRTY READS
FORMAT OVERWRITES
LFP COSTS
14
D. DIRTY READS
FORMAT OVERWRITES
LFP COSTS
WHAT DETERMINES THE IFBL FOR DIRTY READS?
TESTS
FIELD LENGTHS FIELD POSITION NUMBER OF FIELDS IN
FILE SUB/SUPER,SIMPLE DE
SUB, SUPER DESCR FLDS IN FILE
15
D. DIRTY READS
FLD121,FLD262 SUB121,SUB262
FDT
DOES SUB/SUPERDE USE MORE CPU?
FB
FLD001
16
D. DIRTY READS
LFP COSTS
FORMAT OVERWRITES
IFB PROBLEM WITH L3 SHARED GFID
HINT FROM ZAP AN742192 NOTE AT END
IF AN APPLICATION USES THE SAME GLOBAL FORMAT FOR
L3 SEQUENCES CONTROLLED BY DIFFERENT
DESCRIPTORS, THIS COUNTS AS USING DIFFERENT
GLOBAL FORMATS IN PARALLEL.
WHAT IT LOOKS LIKE gt
17
D. DIRTY READS
LFP COSTS
FORMAT OVERWRITES
IFB PROBLEM WITH L3 SHARED GFID
L3
CID CCCC ADDS5 CACA ADDS1 SC FB AP.
DECOMP AP SB (FOR DR CHECK)
DECOMP AP SC (FOR DR CHECK)
LFP
CACA AP , SB
CACA AP , SC
IFB
IFB
18
D. DIRTY READS
LFP COSTS
FORMAT OVERWRITES
IFB PROBLEM WITH L3 SHARED GFID
OP DRES LFP
FOR EXAMPLE, START TEST DB
CID
CMD
FB
ADDS5
ADDS1
L1 AP.
CACA 88

AAAA L2 AP.
CACA 88
BBBB L3 AP.
CACA SB 3112
CCCC L3 AP.
CACA SB 3112
DDDD L3 AP.
CACA SC 6136
19
D. DIRTY READS
LFP COSTS
FORMAT OVERWRITES
IFB PROBLEM WITH L3 SHARED GFID
NATURAL HOW OFTEN DO DIFFERENT L3s SHARE SAME
GFID??
20
D. DIRTY READS
LFP COSTS
FORMAT OVERWRITES
IFB PROBLEM WITH L3 SHARED GFID
NATURAL HOW OFTEN DO DIFFERENT L3s SHARE SAME
GFID??
ADDIT 5 DE COUNT FBL FNR CID
NAT.PGM F8BD49E0A63A909F S8 2 36 142
13700401 GB8OAEN3 F8BD49E0A64AA122 SD 10 72
139 06500401 GB8OAEN4 F8BD49E0A70E5220 SB 1,215
36 48 18300301 GB8OASN3 F8BD49E022BCB486 AC
2 216 139 22800101 GBLCNTP2 F8BD49E022BCB486 SE
24 216 139 21600101 GBLCNTP2 F8BD49E022BCB486 SQ
1,677 216 139 29000201 GBLCNTS7 F8BD49E022BCC886
AB 36 63 137 16950101 GBLCNTP2 F8BD49E0221F09C
0 S2 68 36 255 18500101 GBITMIP1 F8BD49E029D01
E56 SB 22 99 136 10500101 GBUTRBP1 etc
21
D. DIRTY READS
Sx WHO IS AFFECTED (CHECK FOR RSP. 175)
DIRTY READ CHECK MADE FOR Sx ONLY IF

• CMD S1 OR S4
• AND
• FBLgt1 (RETURN 1ST REC)
• AND
• 3. SB 1 Descr
• OR
• SB 1 Descr thru 1 Descr

22
D. DIRTY READS
Sx WHO IS AFFECTED (CHECK FOR RSP. 175)
ALG
CMD
SEARCH BUFFER eg.
CHECK MADE?
23
D. DIRTY READS
Sx WHO IS AFFECTED (CHECK FOR RSP. 175)
FIND EMPLOYEES WITH NAME SMITH
S1 FBL/RBL,IBL4 L1 COPT2N L1 COPT2N L1
COPT2N . . . RC
DIRTY READ CHECK MADE
NO CHECK
NO CHECK
NO CHECK
24
D. DIRTY READS
Sx IFBL FOR IF DIRTY READ CHECK
ANSW SAME AS FOR L3/L6
SUB, SUPER DESCR FLDS IN FILE
25
D. DIRTY READS
S1/L1 DECOMPRESS TO WHERE?
FIND EMPLOYEES WITH SUB-NAME SMITH
26
D. DIRTY READS
S1 CASE STUDY 1
FOR EXAMPLE
FBAB. 1st FDT FIELD DESCR BB SUB-DE OF AC
2ND FDT FIELD VALUEJONES ISNQ1 262 FIELDS IN
FILE IBL4 100K LOOP Sx/RC

• VARY
• S1 SBBB. (DIRTY READ CHECK)
• S1 SBBB,O,BB. (NO D.R. CHECK)
• S1 SBBB,D,BB. (NO D.R. CHECK)
• S2 SBBB. ADDS1BB (NO D.R. CHECK)

27
D. DIRTY READS
S1 CASE STUDY 1, ISNQ1
SUM CMDTIME 100K CALLS
SBBB.
SBBB,O,BB.
SBBB,D,BB.
SBBB. ADDS1BB
CMDS1
CMDS1
CMDS1
CMDS2
28
D. DIRTY READS
S1 CASE STUDY 2, ISNQ20
SUM CMDTIME 100K CALLS
CMDS1
CMDS1
CMDS2
CMDS1
SBBB.
SBBB,O,BB.
SBBB. ADDS1BB
SBBB,D,BB.
29
D. DIRTY READS
SUMMARY WHAT CAUSES MORE CPU USE?
1. USE OF L3/6.
2. USE OF S1/4 (ONLY ALG.1 AND ALG.2 THRU)
3. LAST FB FIELD LESS THAN ADDS1/Sx-SB FIELDS
4. IF SUB/SUPER DE USED (DECOMPRESSES ALL FIELDS
5. IF L3/6 SHARE GFID ACROSS DIFFERENT ADDS1
DESCR.
6. MORE FIELDS IN FILE (esp. SUB/SUPERDE)
7. LFP IS POORLY SIZED (TOO SMALL, TOO BIG)
30
D. DIRTY READS
SUMMARY
CPU DECOMPR. FORMAT - TRANS/FOVWR
INTEGRITY CHECK
VS.
31
D. DIRTY READS
SUMMARY

• 1. LOGIC ON SUB/SUPER DE?
• 2. ZAP FOR S1/4
• ZAP TO DISABLE DIRTY READS?

SAG
??????????
32
D. DIRTY READS
SINCE THIS MATERIAL WAS CREATED.
Spec.ZAP AY743109
PRODUCT ADA743
MODULES ADANC5/ADANC7/ADANC9/ADARUN

PROBLEM PERFORMANCE ZAP WHEN
NUCID0 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
PERFORMENCE NON-CLUSTER SPECIAL
IN THE LAB
L3/L6/S1 IFBL SAME AS V71 NO RSP.CD 175 RETURNED
IF DE-VALUE DOES NOT MATCH DS EFFECTIVELY NEGATES
ABOVE DISCUSSION BUT INTEGRITY CHECK IS
LOST
33
C. QBLKSIZE AGAIN
FOR SEQUENTIAL DATASETS, BY ADAIOR, FOR
UTILITIES (eg. ADASAV, ADAORD)
ADARUN QBLKSIZE up to 256K
BLKSIZE gt 32K ?
OS/390 V2R10 or beyond All IBM PTFs TAPE DRIVE
SUPPORT (256K/3590,STK 9840 64K/3490) OTHER
SOFTWARE SUPPORT SYNCSORT, IDCAMS, . . .
34
C. QBLKSIZE AGAIN
PRODUCTS / ISSUES, 64K,256K
ADASTRIP - WORKS SYNCSORT - WORKS IDCAMS
- FAILS ISSUE MULTIPLE TAPE DEVICE
ENVIRONMENTS
35
C. QBLKSIZE AGAIN
ERROR MESSAGES (IF LARGE BLOCKSIZE NOT SUPPORTED)
IDCAMS - IEC141I 013-E1 SYNCSORT - WER426I SORT
INTERNAL ERROR -- RECOVERY ATTEMPT IN
PROGRESS WRONG DRIVE IEC141I 013-68
36
C. QBLKSIZE AGAIN
IN THE LAB
FOR EXAMPLE
ADASAV, 35 TAPES, 3490 TAPE DRIVE, 64K vs 32K
37
C. QBLKSIZEAGAIN
PRODUCTION DB ME
FOR EXAMPLE
ADASAV, STK 9840 256K vs 32K
(82G)
38
C. QBLKSIZEAGAIN
PRODUCTION DB ME
FOR EXAMPLE
ADASAV, STK 9840 256K vs 32K
(82G)
ELAPSED TIME (MINUTES)
32K
256K
39
C. QBLKSIZEAGAIN
PRODUCTION DB ME
FOR EXAMPLE
ADASAV, STK 9840 256K vs 32K
(82G)
CPU TIME (SECONDS)
32K
256K
40
C. QBLKSIZEAGAIN
PRODUCTION DB NQ
FOR EXAMPLE
ADASAV, STK 9840 256K vs 32K
(84G)
ELAPSED TIME (MINUTES)
32K
256K
41
C. QBLKSIZEAGAIN
PRODUCTION DB NQ
FOR EXAMPLE
ADASAV, STK 9840 256K vs 32K
(84G)
CPU TIME (SECONDS)
32K
256K
42
C.a 32K,256,Virtual Tape
PRODUCTION DB NQ
SYN1 SYN2 TIME (MINUTES)
STK 32K
STK 32K
STK 256K
VIRTUAL 32K