1 ADABAS 7 PERFORMANCE CHECKLIST October, 2001 Jim Poole.

1

ADABAS 7 PERFORMANCE

CHECKLIST

October, 2001Jim Poole

2

TOPICS

A. NEW WITH ADABAS 7.

B. THRUPUT ISSUES

C. CPU ISSUES

D. IO ISSUES

3


WHY FASTER?

1. LBP ACCESS VIA HASHING

10:1 REDUCTION IN INSTRUCTIONS

IN THE LAB: ABOUT 16% CPU IMPROVEMENT.

2. LFIOP ACCESS VIA HASHING.

3. LFIOP AS A BACKUP LBP.

4. BETTER IMPORTANCE CHAIN DETERMINATION.

5. ASYTVS=YES.

4

MEAN CMD TIME (V7: 02/18)

0

0.00001

0.00002

0.00003

0.00004

0.00005

0.00006

0.00007

0.00008

0.00009

0.0001

seco

nds

ADABAS 7 CPU / CMDTIME DIFFERENCES

V7

5


FORWARD INDEX COMPRESSION (FIC)

7. FASTER UTILITIES: ADASAV, ADAORD, ETC.

1. SAVE INDEX SPACE: 46 TO 54%

2. SAVE OVERALL DATABASE SPACE: 10 TO 13 %

3. SAVE ON NEW DASD FOR GROWTH: 10 TO 13 %

4. INCREASE POTENTIAL and EFFICIENCY FOR NI CACHE:

CACHE UI AC (eg. 15K 16K BLOCKS)

CACHE NI ??? (eg. 1700K BLOCKS)

CACHE DS ?? (eg. 400K BLOCKS)5. REDUCE IO FOR L3/L6, L9, Sx:

DATABASE OVERALL

BATCH JOBS DOING LONG L3/L6

6. IMPROVED BUFFER EFFICIENCY.

6

A. NEW WITH ADABAS 7. (FIC) JOB NONAME RUN TIME

0

1000

2000

3000

4000

5000

6000

seco

nds

FIC->

7

A. NEW WITH ADABAS 7. (FIC)

? ANY CPU OVERHEAD ON READS ?

? ANY CPU OVERHEAD ON UPDATES?

! NONE !

FIC EVERYTHING.

8

A. NEW WITH ADABAS 7. ADARES REGENGERATE MTR=YES

0

5

10

15

20

25

30

1

14 27 40 53 66 79 92

105

118

131

144

157

170

183

196

209

222

235

248

261

274

287

300

313

326

339

352

365

378

391

404

417

430

443

456

469

482

495

508

521

534

RUNS

TIM

E (M

INU

TES

)

MTR=NO ->

MTR=YES ->

9

A. NEW WITH ADABAS 7. ONLINE INV, ORD

4. ADARES REGENERATES DATA NICELY.

ONLINE INVERT:1. RUN TIME: SAME AS BATCH (IF NO DATABASE ACTIVITY)

2. USES A TAD MORE NI SPACE THAN BATCH (+2%)

3. DURING ADASAV: WATCH THOSE PLOG IO’S

4. ADARES REGENERATES NICELY.

ONLINE ORD:

1. RUN TIME: 5X BATCH (IF NO DATABASE ACTIVITY)

2. RESPECTS PAD MORE THAN BATCH RESPECTS PAD

3. DURING ADASAV: WATCH THOSE PLOG IO’S

10

TIME 4TIME 3TIME 2TIME 1

B.THRUPUT ISSUES. OVERVIEW.

USER CQ THREAD USER

11

B.THRUPUT ISSUES. TOOLS.

1. ADADBS DCQ

2. ADADBS ECBS (OFF CLOG)

3. THREAD COUNT (SHUTDOWN, DTH, etc)

4. ENQ/DUR (ADARUN CL=4/5)

5. SYNS-5B, ADADBS REFRESHSTATS

6. IS THE LOAD IMPACTING PERFORMANCE?

12

B.THRUPUT ISSUES. TOOLS.TIME 4TIME 3TIME 2TIME 1

USER CQ THREAD USER

1. ADADBS DCQ

X’18’ X’28’

USER JOB FLAG IUBL ----------------------------------------C3C9C3E2C2F1C8E3 (CICSB1HT) CIPRCSBC 28 2927C3C9C3E2C2F1D2D9 (CICSB1KR) CIPRCSBA 28 2927C3C9C3E2C2F1C2D8 (CICSB1BQ) CIPRCSBC 28 367C3C9C3E2C2F1C1F8 (CICSB1A8) CIPRCSBB 18 304B65C11176BA318C3 (:*..,t.C) P1BXK0AR 28 2530C3C9C3E2C2F17CD9 (CICSB1@R) CIPRCSBB 18 2927C3C9C3E2C2F1C2D8 (CICSB1BQ) CIPRCSBC 28 367C3C9C3F7F1F3F4F4 (CIC71344) CIPRCSBH 18 634C3C9C3E2C2F15F7B (CICSB1¬#) CIPRCSBA 18 2927C3C9C3E2C2F1E55E (CICSB1V;) CIPRCSBC 28 380

13


USER CQ THREAD USER

2. ADADBS ECBS (OFF CLOG)

ECB COUNT

MEAN MEAN MEAN MEAN MEAN MAX CMD ENQTIME DURAT CMD PST TOT PST MAX TIME COUNT SECS SECS TIME ECB IO ECB THD------ --------- ------- ------- ------- ---- ---- --- ---165402 291,784 0.00003 0.00063 0.00006 5.8 0.06 24 9165501 1,578,348 0.00003 0.00047 0.00006 5.3 0.03 21 10170001 1,949,641 0.00002 0.00051 0.00005 5.1 0.05 43 15170501 2,171,444 0.00002 0.00038 0.00004 4.3 0.04 20 18171001 2,018,170 0.00001 0.00033 0.00004 3.7 0.04 21 11171501 2,178,302 0.00002 0.00036 0.00004 4.7 0.03 34 14172001 116,406 0.00002 0.00052 0.00005 7.1 0.03 18 8

14


USER CQ THREAD USER

3. THREAD COUNT (SHUTDOWN, DTH, etc)

THRD CNTTHREAD STATUS USAGE------------------------ 1 AA 118574503 2 WI 108594864 3 WI 79809114 4 WI 53275850 5 WI 30081855 6 UU 8260981 7 UU 3245993 8 UU 1740369 9 UU 1146354 10 UU 788271 11 UU 523565 12 UU 287192 13 UU 205306 14 UU 159414 15 UU 126467 16 UU 99034

? 90 TO 95 % OF COMMANDS ?

15


USER CQ THREAD USER

4. ENQ/DUR (ADARUN CL=4/5)

ENQ DURCL=4 ENQ DURCL=5

MEAN MEAN MEAN MEAN MEAN MAX CMD ENQTIME DURAT CMD PST TOT PST MAX TIME COUNT SECS SECS TIME ECB IO ECB THD------ --------- ------- ------- ------- ---- ---- --- ---165402 291,784 0.00003 0.00063 0.00006 5.8 0.06 24 9165501 1,578,348 0.00003 0.00047 0.00006 5.3 0.03 21 10170001 1,949,641 0.00002 0.00051 0.00005 5.1 0.05 43 15170501 2,171,444 0.00002 0.00038 0.00004 4.3 0.04 20 18171001 2,018,170 0.00001 0.00033 0.00004 3.7 0.04 21 11171501 2,178,302 0.00002 0.00036 0.00004 4.7 0.03 34 14172001 116,406 0.00002 0.00052 0.00005 7.1 0.03 18 8

16


5. SYNS-5B, ADADBS REFRESHSTATS

CHECKPOINT CREATED BY:

a. NUCLEUS SHUTDOWN (eg. ADAEND)

b. ADADBS REFRESHSTATS

ALL / CMDUSAGE / COUNTERS / FILEUSAGE

POOLUSAGE / THREADUSAGE

REPORT ON CHECKPOINT:

a. ADAREP CPEXLIST,OFFSET=nn

b. DDPRINT OFF NUC (IDENTICAL TO SHUTDOWN STATS)

17

B.THRUPUT ISSUES. TOOLS. 5. SYNS-5B, REFRESHSTATS

SYNS 5B 2001-09-09 02:35:52 REFRESH STATISTICS THREAD 1: 983037373 COMMANDS THREAD 2: 869948405 COMMANDS THREAD 3: 566715558 COMMANDS : : : : : : REMOTE LOGICAL CALLS 186217341 REMOTE PHYSICAL CALLS 0 LOCAL LOGICAL CALLS 3147407681 LOCAL PHYSICAL CALLS 0 FORMAT TRANSLATIONS 127578298 FORMAT OVERWRITES 42929 AUTORESTARTS 162896 PLOG SWITCHES 145 BUFFER FLUSHES 5353 THROWBACKS ISN PROBS 3536 THROWBACKS SPACE PROBS 0

18

B.THRUPUT ISSUES. TOOLS. 5. SYNS-5B, REFRESHSTATS

HIGH WATER MARKS: COMMAND QUEUE 31296 HOLD QUEUE 112084 FORMAT POOL 2999916 TBI POOL 4288 TBS POOL 76496 USER QUEUE 196444 WORK POOL 636832 SECURITY POOL 0 UQFLIST POOL 53056 UNIQUE DE POOL 17 XID POOL 0 A1/4 COMMANDS 14749218 BT COMMANDS 164487 : : : : :

19


6. IS THE LOAD IMPACTING PERFORMANCE?

SUM OF DURATION ELAPSED TIME

IF > 1, THEN YES, THERE ARE PERFORMANCE IMPACTS.

IF < 1, THEN NO, WHO CARES.

20

B.THRUPUT ISSUES. SOLUTIONS. 1. WAIT-ON-HOLD. 2. COMPLEX COMMAND WAIT. 3. UPDATE COMMAND WAIT. 4. CLOG IO. 5. UEX4 AND COMMAND LOG TOOL. 6. EXCESSIVE COMMAND LOG REPORTING. 7. BAD UEX1/2. 8. PATH LENGTH. 9. SYSTEM PRIORITIES.10. EXCESSIVE BATCH JOBS.11. EXCESSIVE MULTI/PREFETCH JOBS.12. EXCESSIVE BAD SEARCHES.13. EXCESSIVE NON-DESCRIPTOR SEARCHES.14. EXCESSIVE SX/L3 ON EXPANDED FILES.15. USER WORKLOAD IMPROVEMENTS.16. NEW DATABASE VS. SMP.17. SEE CPU ISSUES.18. SEE IO ISSUES.

21

B.THRUPUT ISSUES. SOLUTIONS. 1. WAIT-ON-HOLD.

L4,L5,L6,S4,HI vs. L1,L2,L3,S1

HOLDERS: SUMMARIZE(SUBTOT=OFF) COUNT MEAN(ENQ-TIME) (PF=N1.5) MEAN(DURATION) (PF=N2.5) MAX(DURATION) MEAN(CMD-TIME) (PF=N1.5) MEAN(TOTAL-IO) (PF=N1.2) MEAN(ECBS) (PF=N2.2) BY CMD*** INTERVAL N HOURS WHERE CMD(1) = 'L' TITLE = ‘HOLD VS NOT HOLD’ INSIGHT-LINES = 100** PRINT TO SUMM2 ;

22


ANALYSYS OF HOLD OPTION

C MEAN MEAN MAX MEAN MEAN MEAN M CMD ENQTIME DURAT DURAT CMD TOT PST D COUNT SECS SECS SECS TIME IO ECB -- ------- ------- ------- ------- ------- ---- ----- L1 200,659 0.00003 0.00037 0.286 0.00003 0.04 5.07 L4 3,286 0.00003 0.43218 9.456 0.00007 0.03 6.58

L2 204,753 0.00003 0.00022 0.061 0.00003 0.02 4.19

L3 542,935 0.00003 0.00056 0.672 0.00006 0.05 5.31 L6 1,296 0.00003 0.00268 0.138 0.00026 0.21 6.22

23

B.THRUPUT ISSUES. SOLUTIONS. 1. WAIT-ON-HOLD. LONGHOLD: EXTRACT(ML=500000) TIME* JOBNAME TP-TRANS-NAME (HD='TRAN','NAME')* PROGRAM NATURAL-LOGON NAT-PROGRAM CID (PF=H8) FILE (PF=N3) COMMAND-CODE COPT1 DURATION (PF=N3.3)* ENQ-TIME (PF=N1.5) CMD-TIME (PF=N1.5) ISN (PF=N8) TOTAL-IO (PF=N4) WHERE (CMD = 'L4' OR CMD = 'L6‘) AND DURATION > 5 INSIGHT-LINES = 300** PRINT TO ADHOC2;

24


C C TRAN NATURAL FIL M O DURAT CMD TIME NAME PROGRAM CID NUM D 1 SECS TIME110948 NATP LQNWCTN2 47200201 95 L4 . 9.456 0.00016

TOT ISN IO1,973,382 0

THROWBACKS ISN PROBS 3536

NUC SHUTDOWN DDPRINT or SYNS-5B CHECKPOINT

MEASURED OVER TIME.

25

B.THRUPUT ISSUES. SOLUTIONS. 1. WAIT-ON-HOLD. SOLUTIONS:

1. CB COP1=‘R’

RSP CD 145

2. NATURAL WH=OFF (SECURITY), TRAP FOR NON-0 RSP CD. NOTROPT: SUMMARIZE(SUBTOT=OFF)

COUNT MIN(CID) (PF=H8) MAX(CID) (PF=H8) MIN(FILE) (PF=N3) MAX(FILE) (PF=N3) BY NAT-PROGRAM WHERE (CMD = 'L4' OR CMD = 'L6') AND COPT1 NE 'R' AND NAT-PROGRAM > ' '* AND CID NE H'00000000' INSIGHT-LINES = 100** PRINT TO SUMM2;

26


MIN MAX NATURAL CMD MIN MAX FIL FIL PROGRAM COUNT CID CID NUM NUM-------- ---------- -------- -------- --- ---LQPBKY18 1 43050101 43050101 182 182LQCCVDN1 36 07500201 10900201 76 78LQVENKP3 13 07500201 07500201 182 182LQVXAPP1 23 33450101 34100101 131 169

? WHEN IS ‘R’ OPTION NOT GOOD?

? BATCH ?

? BACKGROUND ?

? ?

27

B.THRUPUT ISSUES. SOLUTIONS. 2. COMPLEX COMMAND WAIT.

USER CQ THREAD?

COMPLEX CMD IF :

1. CMD = S2, S8, S92. SBL > 9 COMPLEX COMMANDS NOT SELECTED

FROM CQ TO THREAD IF:

1. LESS THAN 50% WORK-2 AVAILABLE.2. LESS THAN LS AVAILABLE IN LWP3. AT LEAST ONE COMMAND WAITING ON LWP SPACE

28


CMDTYPE: SUMMARIZE (SUBTOT=ON) COUNT MEAN(ENQ-TIME) (PF=N1.5) MEAN(DURATION) (PF=N1.5) MEAN(CMD-TIME) (PF=N1.5) MEAN(ECBS) (PF=N2.1) MEAN(TOTAL-IO) (PF=N1.2)** BY JOBNAME BY COMMAND-TYPE INTERVAL 15 MINUTES INSIGHT-LINES 100** PRINT TO SUMM1 ;

29

B.THRUPUT ISSUES. SOLUTIONS. 2. COMPLEX COMMAND WAIT. T MEAN MEAN MEAN MEAN MEAN

Y CMD ENQTIME DURAT CMD PST TOTP COUNT SECS SECS TIME ECB IOC 435 0.00002 0.00139 0.00013 7.8 0.04S 335,188 0.00002 0.00106 0.00005 5.4 0.08U 28,500 0.00002 0.00643 0.00050 8.6 0.44

1. CL=5

T MEAN MEAN MEAN MEAN MEANY CMD ENQTIME DURAT CMD PST TOTP COUNT SECS SECS TIME ECB IOC 435 0.00002 0.54339 0.00013 7.8 0.04S 335,188 0.00002 0.00106 0.00005 5.4 0.08U 28,500 0.00002 0.00643 0.00050 8.6 0.44

2. CL=5


3. CL=4

30


THROWBACKS SPACE PROBS 190

NUC SHUTDOWN DDPRINT or SYNS-5B CHECKPOINT

31

B.THRUPUT ISSUES. SOLUTIONS. 2. COMPLEX COMMAND WAIT. a. WORK-2 SOLUTION

ADARUN LWKP2= ! USE !

MIN (<16M REC FILES): 24,000 (3390) 10,000 (8391)

LWKP2 = ( S x 2 x TOPISN ) / ( ( WB – 16) / 4 )

WB = 5724 or 1368216 = BLOCK HEADER4 = 4b PER ISN TOPISN = OF LARGEST FILE FOR SX CMDS 2 = ALWAYS ENSURE 50% AVAILABLE S = PEAK NO. OF CONCURRENT COMPLEX COMMANDS. %Sx TIMES USED THREADS (?TIMES 2?)

32

B.THRUPUT ISSUES. SOLUTIONS. 2. COMPLEX COMMAND WAIT b. WORK-POOL SOLUTION

ADARUN LWP/LS=

LWP = LS * 3

= LS + (HI-WATER BYTES FROM DDPRINT)

= LS * S (WHERE S = 2 * (%Sx) * NT)

= LS * NT

33

B.THRUPUT ISSUES. SOLUTIONS. 3. UPDATE COMMAND WAIT.

USER

UPDATE CMD IF :

CMD = A1,A4,E1,E4, N1,N2, ET,BT,OP,CL

UPDATE COMMANDS NOT SELECTEDFROM CQ TO THREAD IF:

1. LFIOP=0 AND BUFFER FLUSH, WORK WRITE, OR PLOG WRITE.

2. ET-SYNC: ADAEND ADASAV SYN2 OP CMD SYNCC

CQ THREAD?

34



1. CL=5


2. CL=4

35


a. LFIOP SOLUTION

ADARUN LFIOP= 1 or n ! USE !

b. ADASAV SOLUTION

NOTE UPDATE LOAD VERSUS ADASAV TIME

ADASAV TTSYN=120 ?

36


* USE ADARUN CL=4MCISYN2: SUMMARIZE COUNT MEAN(ENQ-TIME)**** MIN(ENQ-TIME)**** MAX(ENQ-TIME) MIN(TIME) MAX(TIME) BY ENQ-TIME (BY 5) WHERE JOBNAME2(2) = 'CI' AND TIME GE '204500‘ AND TIME LE '235959' AND ENQ-TIME LE 240 INTERVAL 15 MINUTE TITLE='TIME TO SYN2 TO ET-SYNC' PRINT TO ADHOC1;

37

B.THRUPUT ISSUES. SOLUTIONS.

4. CLOG IO.

! ELIMINATE OR REDUCE !

EACH CLOG IO IS A WAIT IO !

DATABASE XYZFULL COMMAND LOGGING200 COMMANDS PER SECMEAN ECBS = 11.3

NOLOGGING

MEAN ECBS = 7.0

1400 CMDS PER SECFULL CLOGGINGCLOG BLKSIZE = 5064Cloglayout = 5LRECL = 172RECS PER BLK = 29IO’S PER SEC = 48.3MEAN IO TIME = .00448.3 X .004 = .19 SEC

.19 / 1 = 19% WAITTIME

38


5. UEX4 AND COMMAND LOG TOOL.

HOMEGROWN

TRIM: PART OF NUC

REVIEW: SUBTASK OR HUB

APAS/INSIGHT: SUBTASK

39


6. EXCESSIVE COMMAND LOG REPORTING.

???EVERY HOUR,DB EXPERIENCINGSLOWDOWN ??? SUMMARY REPORT:

COUNT, DURATION,ETC. BY JOBNAME BY NAT-PROGRAM BY FILE BY CMD INTERVAL 1 HOUR

40


7. BAD UEX1/2.

DATABASE IS SINGLE-THREADED AT THIS TIME.

41


8. PATH LENGTH. SAME LPAR, DIFFERENT LPAR, REMOTE

MEAN MEAN MEAN MEAN MEAN CMD ENQTIME DURAT CMD PST TOTJOB NAME COUNT SECS SECS TIME ECB IO-------- ------ ------- ------- ------- ---- ----CICSPROD 1,416 0.00002 0.00152 0.00027 2.0 0.89DB00SAME 1,002 0.00003 0.00006 0.00001 1.9 0.02DB00LPAR 1,002 0.00081 0.00009 0.00002 1.9 0.02DB00VTAM 1,002 0.18632 0.00014 0.00002 2.4 0.05

42


9. SYSTEM PRIORITIES.

1. ADABAS

2. TP

3. BATCH

PRE - WLM

1. VELOCITY HIGH

2. SYSSTC

WLM

? CPU WAIT ?

? ENQ-TIME ?

? CMD-TIME ?

43


10. EXCESSIVE BATCH JOBS.

a. EACH BATCH JOB TENDS TO DOMINATE ONE MORE

THREAD

11. EXCESSIVE MULTI/PREFETCH JOBS. a. SHOULD RUN NON-PEAK TIMES.

b. SHOULD RUN FEWER THAN 2-3 CONCURRENTLY.

44

B.THRUPUT ISSUES. SOLUTIONS. 10,11. EXCESSIVE JOBS.

MEAN ECBS VS BATCH JOBS

0

5

10

15

20

25

30

35

1 8

15 22 29 36 43 50 57 64 71 78 85 92 99

106

113

120

127

134

141

148

155

162

169

176

183

190

197

204

211

218

225

232

239

246

253

260

267

274

281

ECBS COUNTECBS BATCH JOBS

45


DECLARE FETCH (A3) VALOF IF COPT1 = 'M' RESULTIS 'YES' ELSEIF COPT1 = 'P' RESULTIS 'YES' ELSEIF COPT1 = 'O' RESULTIS 'YES' ELSE RESULTIS ' ' IFEND;DBUSE: SUMMARIZE(GTT=YES) COUNT MEAN(ENQ-TIME) (PF=N1.5) MEAN(DURATION) (PF=N1.5) MEAN(CMD-TIME) (PF=N1.5) MEAN(ECBS) (PF=N2.1) MEAN(TOTAL-IO) (PF=N1.2) MAX(FETCH) (HD='FTCH') BY JOBNAME INTERVAL 15 MINUTES INSIGHT-LINES=300;

46

B.THRUPUT ISSUES. SOLUTIONS. 10,11. EXCESSIVE JOBS. MEAN MEAN MEAN MEAN MEAN CMD ENQTIME DURAT CMD PST TOTJOB NAME COUNT SECS SECS TIME ECB IO FTCHCICSPRDA 39 0.00002 0.00045 0.00012 10.4 0.18CICSPRDB 2,217 0.00003 0.00037 0.00007 9.3 0.04CICSPRDC 10,231 0.00002 0.00093 0.00006 9.2 0.14CICSPRDD 524 0.00003 0.00029 0.00006 9.3 0.04CICSPRDE 1,055 0.00003 0.00070 0.00009 9.7 0.10CICSPRDF 274 0.00002 0.00041 0.00005 9.7 0.02CICSPRDG 2,246 0.00002 0.00172 0.00021 8.9 0.34CICSPRDH 311 0.00003 0.00059 0.00005 8.9 0.05CICSPRDU 12 0.00118 0.00060 0.00007 10.8 0.17JOB00001 6,660 0.00000 0.00104 0.00005 8.5 0.18JOB00002 5,815 0.00000 0.00118 0.00004 8.4 0.07JOB00003 5,497 0.00000 0.00130 0.00004 8.4 0.09JOB00004 3,892 0.00000 0.00032 0.00003 9.4 0.02JOB00005 3,127 0.00000 0.00031 0.00002 9.5 0.02JOB00006 15,833 0.00000 0.00027 0.00002 8.5 0.03 ======== 57,733 0.00001 0.00074 0.00005 8.8 0.09

47


MEAN MEAN MEAN MEAN MEAN CMD ENQTIME DURAT CMD PST TOTJOB NAME COUNT SECS SECS TIME ECB IO FTCH-------- ------- ------- ------- ------- ---- ---- ----CICSPRD1 5,261 0.00003 0.00073 0.00008 3.7 0.13CICSPRD2 494 0.00003 0.00030 0.00011 4.2 0.03CICSPRD3 267 0.00003 0.00095 0.00009 3.1 0.21CICSPRD4 233 0.00003 0.00212 0.00009 4.5 0.22DDBTCHA1 5,158 0.00002 0.00109 0.00016 4.6 0.09DDBTCHB6 8 0.00002 0.00020 0.00006 5.5 0.00DDBTCHHQ 4,407 0.00003 0.00200 0.00015 3.6 1.11DDBTCHXY 226 0.00003 0.06774 0.00113 7.7 5.65 YES

48

B.THRUPUT ISSUES. SOLUTIONS. 12. EXCESSIVE BAD SEARCHES.

FIND CLIENTS WITH CITY=‘NEW YORK’ AND NAME=‘SMITH’ AND AGE>21 AND WEIGHT=321

49

B.THRUPUT ISSUES. SOLUTIONS. 12. BAD SEARCHES.

HIGHIO:SUMMARIZE(SUBTOT=OFF) COUNT MEAN(TOTAL-IO) MEAN(DATA-IO) MEAN(ISNQ) MEAN(DURATION) BY JOBNAME BY NAT-PROGRAM BY CID (PF=H8) BY CMD WHERE TOTAL-IO GT 100 OR DURATION GT 1.0

* PRINT TO SUMBAD INSIGHT-LINES 300;

HIGHIOE: EXTRACT(ML=1000000) DATE-TIME COMMAND-CODE FILE JOBNAME CID (PF=H8) TP-TRANS-NAME NAT-LOGON NAT-PROGRAM PROGRAM CICS-TERMID ISNQ TOTAL-IO DURATION WHERE . . . .

50

B.THRUPUT ISSUES. SOLUTIONS. 12. BAD SEARCHES.

C MEAN MEAN JOB/ M CMD TOT DATA TRANS PROGRAM CID D COUNT IO IO-------- -------- -------- -- ----- -------------NATO XYVADDP1 01300101 S1 3 178.3 0.0

MEAN MEAN ISN DURAT QUANT SECS-------- -------29,527.0 0.4050

51

B.THRUPUT ISSUES. SOLUTIONS. 13. EXCESSIVE NON-DESCRIPTOR SEARCHES.

a. REPORT ON VIA: WHERE CMD(1) = ‘S’ AND DATA-IO>1

b. STOP VIA: ADARUN NONDES=NO

52

B.THRUPUT ISSUES. SOLUTIONS. 14. EXCESSIVE SX/L3 ON EXPANDED FILES.

EXPANDED FILE 4-BYTE ISN FILEvs.

MANAGEABILITY vs PERFORMANCE

SPACE SAVINGS vs SPACE SAVINGS

(3B vs 4B ISN IN ASSO) (NO DUPE DE-VALUE

ACROSS FILES)

! UNLESS MANAGEABILITY IS AN ISSUE, USE 4-BYTE ISN FILES !

(vs EXPANDED FILES)

53

B.THRUPUT ISSUES. SOLUTIONS. 14. EXCESS SX/L3, EXP.FILES.

Sx : Eg. NAME=BOB

NON-EXP. FILE

FILE-X NI :

BOB 1,5,8

1,5,8,9,10,11R.I.L.(Resultant Isn List)

EACH NI BLOCK: MADE UP OF NI ELEMENTS.

EACH NI ELEMENT: DE-VALUE / COUNT / ISN LIST

541,5,8,9,10,11R.I.L

.(Resultant Isn List)


Sx : Eg. NAME=BOB

EXP. FILE (4 COMPONENTS)

FILE-W NI :

BOB 1

FILE-X NI :

BOB 5,8

FILE-Y NI :

BOB 9

FILE-Z NI :

BOB 10,11

55


L3/6 THRU FILE

NON-EXP. FILE

FILE-X NI :

BART BOB,BRADBILLBEA,BETH

L3’S: 1 2 3 4 5 6

NOTE: 1 TBQ(LQ=) ELEMENT CREATED AND USED.

56


EXP. FILE (4 COMPONENTS)

FILE-W NI :

BART BUDBOBBETH

L3/6 THRU FILE

NOTE: 4 TBQ(LQ=) ELEMENTS CREATED AND USED.

FILE-X NI :

BART . . .BETHBEA

FILE-Y NI :

BETH . . .BOBBILL

FILE-Z NI :

BEA . . .BRADBOB

L3’S: FOLLOW THE NUMBERS

1

2 3

4

5

6

7 8

9

10

11 12

13

57


CUSTOMER EXPERIENCE (SAG-L)

BEFORE: 29 COMPONENT FILES, 3B ISN, 16M RECS @

AFTER: 10 COMPONENT FILES, 4B ISN, 50M RECS @

BEFORE: L3’S DOING UP TO 60 IO’S AFTER: L3’S DOING UNDER 10 IO’S

BEFORE: BATCH JOB RUNS IN 11 HOURSAFTER: BATCH JOB RUNS IN 3 HOURS

58

B.THRUPUT ISSUES. SOLUTIONS. 15. USER WORKLOAD IMPROVEMENTS. a. HISTORICAL – PERIODIC REVIEW OF THE TOP 10% b. ONLINE

BY JOBNAMEBY TP-TRANS-NAMEBY PROGRAMBY NAT-PROGRAM

WHO PEAK HOURSDAILYWEEKLY

WHEN

COUNTDURATIONCMD-TIME TOTAL-IO

WHAT

BY NAT-PROGRAMWHERE JOBNAME(3)=‘CIC’ AND NAT-PROGRAM > ‘ ‘

INTERVAL 168 HOURSWHERE TIME GE ‘090000’ AND TIME LE ‘170000’

59

B.THRUPUT ISSUES. SOLUTIONS. 15. USER WORKLOAD.

PROGRAM COUNT DURATION TOTALIO CMDTIMELQNSCTNH 33,455,949 5,888.379 786 826.723LQ2DCTN1 25,876,640 7,441.052 396,146 1,228.789LQLWNTN1 23,581,968 10,270.808 880,891 791.377GGLGNUP2 9,548,540 2,392.155 26,559 498.448LQLGLCA1 7,254,208 9,371.013 1,248,313 495.574GGLCNEP1 6,529,145 1,902.695 49,985 352.686LQ24C2N3 6,420,988 1,472.643 15,476 291.566LQNWDGF2 5,486,722 2,103.440 57,322 382.099LQLCNTP1 5,270,677 1,802.396 46,019 412.284SECUPTER 4,956,576 1,227.651 1,473 351.807

SORT BY: COUNT , DURATION , TOTALIO , CMDTIME

REVIEW USER BY: (CID), FILE, CMD

a. HISTORICAL – PERIODIC REVIEW OF THE TOP 10

60

B.THRUPUT ISSUES. SOLUTIONS. 15. USER WORKLOAD.b. ONLINE

QUICK: SUMMARIZE COUNT MEAN(ENQ-TIME) MEAN(DURATION) MEAN(CMD-TIME) MEAN(ECBS) MEAN(TOTAL-IO) INTERVAL 10 SECONDS BY JOBNAME** BY TP-TRANS-NAME* WHERE JOBNAME = 'CICSPRDA'** BY GLOBAL-USER* WHERE JOBNAME = 'CICSPRDA'* AND TP-TRANS-NAME = 'NATP' INSIGHT-LINES=300;

1

2

3

61


MEAN MEAN MEAN MEAN MEAN CMD ENQTIME DURATION CMD PST TOTJOB NAME COUNT SECS SECONDS TIME ECB IO-------- ------ -------- -------- ------- ---- ----CICSPRDA 31,069 0.000021 0.00025 0.00005 3.15 0.01CICSPRDB 19,533 0.000023 0.00021 0.00005 3.24 0.00CICSPRDC 3,639 0.000022 0.00046 0.00005 3.79 0.03CICSPRDD 520 0.000023 0.00089 0.00007 3.63 0.09CICSPRDE 41 0.000016 0.00107 0.00013 4.68 0.12CICSPRDF 5,224 0.000020 0.00055 0.00009 3.51 0.05JOBWXYZ 1,201 0.000033 0.00049 0.00005 4.00 0.02JOBZYXW 52 0.000880 0.00022 0.00012 3.50 0.50 ------ -------- -------- ------- ---- ----======== 61,341 0.000022 0.00029 0.00005 3.27 0.02

BY JOBNAME

62


TP MEAN MEAN MEAN MEAN MEANTRANS CMD ENQTIME DURATION CMD PST TOT NAME COUNT SECS SECONDS TIME ECB IO----- ------ -------- -------- ------- ----- -----NATP 14,411 0.000022 0.00024 0.00006 2.06 0.03NATX 123 0.000024 0.00018 0.00007 1.41 0.02COB1 963 0.000023 0.00045 0.00004 2.54 0.04COB2 4,072 0.000023 0.00020 0.00009 2.39 0.00 : : : : : : : : : : :

BY TP-TRANS-NAME

63


MEAN MEAN MEAN MEAN MEANTERM OR CMD ENQTIME DURATION CMD PST TOTTRAN ID COUNT SECS SECONDS TIME ECB IO------- ----- --------- --------- -------- ----- ----CICSD2.K 1407 0.000006 0.00002 0.00002 6.14 0.00CIC92111 7 0.000011 0.00019 0.00005 3.43 0.00CICSD1$: 7 0.000011 0.00018 0.00005 2.86 0.00CICSD2.K 5 0.000012 0.00079 0.00008 2.60 0.20

BY GLOBAL-USER

64


WHODID: EXTRACT TIME** JOBNAME TP-TRANS-NAME** NAT-LOGON NAT-PROGRAM (HD='PROGRAM') CID (PF=H8) CMD FILE (PF=N3) ISN (PF=N8) RESPONSE-CODE (HD='RSP','CD ') DURATION TOTAL-IO (PF=N2) L3-SEQUENCE (PF=A2) COPT1 COPT2 ISNQ (PF=N4) SUB-CODE (HD='RSP','SUB') WHERE GLOBAL-USER = ' ' INSIGHT-LINES = 300;

65


C C C M FIL RSP DURAT TOT L3 O O TIME PROGRAM CID D NUM ISN CD SECS IO SEQ 1 2------ -------- -------- -- --- ------- --- ----- --- --- - -175059 NATPXTNB 07500301 L3 21 114,512 0 0.000 0 TS . V175059 NATPXTNB 07500301 RC 0 0 0 0.000 0 .. S I175059 NATPXTNB 07500301 L3 21 114,512 0 0.000 0 TS . V175059 NATPXTNB 07500301 RC 0 0 0 0.000 0 .. S I175059 NATPXTNB 07500301 L3 21 114,512 0 0.000 0 TS . V175059 NATPXTNB 07500301 RC 0 0 0 0.000 0 .. S I175059 NATPXTNB 07500301 L3 21 114,512 0 0.000 0 TS . V175059 NATPXTNB 07500301 RC 0 0 0 0.000 0 .. S I175059 NATPXTNB 07500301 L3 21 114,512 0 0.000 0 TS . V175059 NATPXTNB 07500301 RC 0 0 0 0.000 0 .. S I175059 NATPXTNB 07500301 L3 21 114,512 0 0.000 0 TS . V175059 NATPXTNB 07500301 RC 0 0 0 0.000 0 .. S I: : : : : : : : :

66

B.THRUPUT ISSUES. SOLUTIONS. 16. NEW DATABASE VS. SMP.

THRUPUT

WHAT SMP SOLVES: WHAT SMP DOES NOT SOLVE:

1. CPU (10-15% INCREASE)

2. IO (LOSS OF CACHING

200% INCREASE) 3. SVC DEPENDENCY

(CONVERSIONS)

67

B.THRUPUT ISSUES. SOLUTIONS. 16. NEW DATABASE VS. SMP.

POINTS TO CONSIDER BEFORE SMP:

- OTHER ADABAS TUNING (THIS DOC)

- CURRENT / PLANNED CPU UPGRADES

- MAX OF CURRENT ARCHITECTURE:

CMDS / SEC vs. CMDS / SEC OF CPU

CMDS / SEC OF CPU = 1 / (CPU SECS / CMD)

( FOR CPU SECONDS PER COMMAND, SEE NEXT SECTION )

68


17. SEE CPU ISSUES.18. SEE IO ISSUES.

69

B. CPU ISSUES. TOOLS.

1.DDPRINT FROM NUC SHUTDOWN.

2.DSTAT AND IOF.

3.ADAREP SYNS-60 CHECKPOINT.

4.CLOG COMMAND TIME.

5.STROBE,OMEGAMON,etc.

70

B. CPU ISSUES. TOOLS. 1.DDPRINT FROM NUC SHUTDOWN.

DURATION 00296:30:07 HOURSWAIT-TIME 00205:17:52 HOURSCPU-TIME 00105:53:32 HOURS

= 1,067,407 SECONDS

= 381,212 SECONDS

CPU = 35.7 %

A D A B A S EXECUTED 961341,857 CALLS

THREAD NUMBER 1 983037,373 2 869948,405 : : : : TOTAL 961341,857

CORRECTED= 3,333,625,4390.00011 CPU SECS/CMD

71

B. CPU ISSUES. TOOLS. 1.DDPRINT FROM NUC SHUTDOWN.

DATE DB TOTAL CMDS HOURS TOTALIO CM/S IO/CM CPU/CM20010715 099 3045887222 286.6 275519717 2952 0.09 .0001220010727 099 2438702234 215.3 216822796 3146 0.09 .0001220010805 099 3440990001 307.9 317241601 3104 0.09 .0001220010827 099 3333625439 296.5 287554044 3123 0.08 .00011

DDPRINT HISTORY

72

B. CPU ISSUES. TOOLS. 2.DSTAT AND IOF.

JOBNAME--JOBID--SYID----CPU-----I/O----STEP-----PROCSTEPDB099NUC S20260 PRD2 21:05:50 7281060 DB099NUC NUCSTEP

TSO ISPF IOF

= 75,950 SECONDS

ADAN17 0099 2001-09-11 12:38:59 NR. OF COMMANDS=595042096

NUC SYSLOG OF DSTAT

= 595,042,096

0.00013 CPU SECS/CMD

73

B. CPU ISSUES. TOOLS. 3.ADAREP SYNS-60 CHECKPOINT.ADARUN INTNAS=3600 or 900ADAREP CPEXLIST

SYNS 60 2001-09-11 12:14:19 718 204371 DUAL ADABAS NUCLEUS STATISTIC NR OF CALLS = 12955548 NR OF PHYSICAL IO = 1422790 NR OF LOGICAL IO = 42672537 TIME = 2001-09-11 12:14:19 CPU USED = 00:28:33

= 1,713 SECONDS= 12,955,548 CMDS

0.00013 CPU SECS/CMD

74

B. CPU ISSUES. TOOLS. 4.CLOG COMMAND TIME.ALSO “ADABAS THREAD TIME”DEFN.

UNITS

TIME IN THREAD MINUS WAITS: IO, OVERLAP

CONTROL BLOCK: 16 MICROSECONDS, 0.000016

C MEAN MEAN MEAN M CMD TOT DURAT CMD D COUNT IO SECS TIMEA1 1,558 0.24 0.00060 0.00026BT 71 0.00 0.00012 0.00001CL 1,759 0.00 0.00026 0.00000ET 1,526 0.00 0.00016 0.00001L1 62,993 0.00 0.00013 0.00005L2 7,793 0.02 0.00026 0.00005L3 171,553 0.86 0.00164 0.00011L4 722 0.00 0.00035 0.00025L6 450 0.15 0.00082 0.00025L9 4,250 0.02 0.00021 0.00007N1 304 3.05 0.00724 0.00105RC 83,828 0.00 0.00010 0.00001: : : : : : : : == 385,841 0.42 0.00089 0.00008

C MEAN MEAN MEAN M CMD TOT DURAT CMD D COUNT IO SECS TIMEA1 1,759 0.00 0.00035 0.00025BT 86 0.00 0.00013 0.00001CL 1,095 0.00 0.00014 0.00000ET 1,857 0.00 0.00015 0.00001L1 82,262 0.00 0.00012 0.00005L2 12,293 0.00 0.00016 0.00004L3 121,410 0.00 0.00014 0.00006L4 919 0.00 0.00037 0.00025L6 541 0.00 0.00034 0.00024L9 5,467 0.00 0.00018 0.00007N1 88 0.00 0.00124 0.00105RC 110,543 0.00 0.00009 0.00001: : : : : : : :== 385,119 0.00 0.00013 0.00005

75

B. CPU ISSUES. TOOLS. 4.CLOG COMMAND TIME.

0

100

200

300

400

500

600

700

1 3 5 7 9

11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83

CMDTIME CPUSECS

76

B. CPU ISSUES. TOOLS. 5.STROBE,OMEGAMON,etc. ADDRESSES OF HIGH CPU

| LMOD Addr | LMOD Name |% of Samples| 000E10E0 | ADAMPM | 71.60 | 00FCA000 | IEANUC01 | 20.63 | 000BC970 | ADANC8 | 2.39 | 00011428 | ADAIOS | 1.44 | 000ADC00 | ADANC7 | 1.24 | 00000000 | *-UNKN-* | 1.17 | 0004FC00 | ADANC1 | .84 | 0003E2C0 | ADACSH | .66

OMEGAMON-CPU-NUC-CPU-ADARUN:

| Address | Offset |% of Samples| 000C21D0 | + 005860 | .00 || 000C2820 | + 005EB0 | .00 || 000C2E70 | + 006500 | .00 || 000C34C0 | + 006B50 | .00 || 000C3B10 | + 0071A0 | .00 || 000C4160 | + 0077F0 | 92.72 || 000C47B0 | + 007E40 | .00 |+____________+__________+____________+

ADANC8

77

B. CPU. SOLUTIONS.

1. BAD Sx COMMANDS. 2. BAD NON-DE SEARCHES. 3. LOOPING PROGRAMS. 4. HIGH CPU USERS. 5. OVER REPORTING WITH CLOG TOOLS. 6. ADARUN LBP. 7. ADARUN LFP, CID’S. 8. ADARUN NC. 9. ADARUN NU,NT.10.DVT CONSTRUCTION.11.COMPRESSION/DECOMPRESSION.12.BAD CACHE PARMS.13.ADABAS RELEASE.14.CPU HARDWARE.

78

B. CPU. SOLUTIONS.

SEE A.THRUPUT SOLUTIONS: 12. EXCESSIVE BAD SEARCHES.1. BAD Sx COMMANDS.

2. BAD NON-DE SEARCHES.SEE A.THRUPUT SOLUTIONS: 13. EXCESSIVE NON-DE SEARCHES.

3. LOOPING PROGRAMS.

4. HIGH CPU USERS.

5. OVER REPORTING WITH CLOG TOOLS.

SEE A.THRUPUT SOLUTIONS: 15. USER WORKLOAD IMPROVEMENTS.

SEE A.THRUPUT SOLUTIONS: 15. USER WORKLOAD IMPROVEMENTS.

SEE A.THRUPUT SOLUTIONS: 6. EXCESSIVE CMD LOG REPORTING.

79

B. CPU. SOLUTIONS. 6. ADARUN LBP.! COST OF BUFFER FLUSHES !

LBP=100M HEADERS = 100M / 2544 = 39K

LAB TEST:CYCLE NUC BETWEEN RUNSFILL LBP WITH X BLOCKSPGM: LOOP (10K): A1 SAME FILE/ISN NO ET C1 (COPT=F)

TEST BLKS/HDRS AVER.NUC RANGE OFRUNS LBP= IN LBP CPU (SECS) NUC CPU---- ---- --------- ----------- -------- 4 150M <20 19 SECS 18 TO 22 6 150M >20K 43 SECS 39 TO 46 4 200M >20K 40 SECS 40 TO 41 3 200M >36.7K 62 SECS 61 TO 63

THE RESULTS:

80

B. CPU. SOLUTIONS. 6. ADARUN LBP. BUFFER FLUSHES.MEASUR

E:

=FLUSH EVERY 6.8 SECSDURATION 00166:30:07 HOURS88,355 BUFFERFLUSHES WERE DONE

1. SHUTDOWN DDPRINT or SYNS-5B

Physical Reads Writes Call DistributionASSO 5341459 844025 Remote Logical ......... 56620167DATA 60343074 274962 Remote Physical ........ 0WORK 293117 4317457 Local Logical ......... 573248391PLOG 1905450 Local Physical ........ 0Logical Reads ....... 0 Logical Reads (binary) . 7FFFFFFFBuffer Efficiency ... 38.0 No. of HQEs active ..... 7 No. of UQEs in User Queue. 296Format Translations .44825232 No. of CQEs waiting in CQ . 0Format Overwrites ... 3818 Total intern. Autorestarts . 8230Throw Backs for ISN . 344 No. of PLOG switches ....... 39Throw Backs for Space. 0 No. of Bufferflushes ...... 1279

2. AOS . . . SYSTEM STATUS

81

B. CPU. SOLUTIONS. 6. ADARUN LBP. BUFFER FLUSHES.MEASUR

E:BUFFER: SUMMARIZE MIN(DATE) MIN(TIME) COUNT SUM(TOTAL-IO) BUFFER-FLUSHES-INTERVAL BUFFER-FLUSHES-SESSION INTERVAL 15 MINUTES INSIGHT-LINES=300;

1. APAS INSIGHT.

SUM BUFF BUFF MIN MIN CMD TOT FLUSH FLUSH DATE TIME COUNT IO INTER SESS010910 154500 4,701,761 199,258 8 754010910 160000 4,663,621 242,943 10 764010910 161500 3,320,363 256,838 10 774010910 163000 3,506,226 223,657 12 786

82

B. CPU. SOLUTIONS. 6. ADARUN LBP. BUFFER FLUSHES.

a. MINIMIZE BUFFER FLUSHES:

ADARUN LP=65,530

? LARGER WORK BLKSIZE ?

ADARUN LFIOP=< ¼ LBP

>AVER. WRITE IO’SEG. 4M TO 5M

b. MONITOR LBP SIZE:

ADARUN LBP=40-60MUSE ADABAS CACHING

83

B. CPU. SOLUTIONS. 7. ADARUN LFP, CID’S.

FB=AA,020,A,..

PROGRAM TRANSLATE

IFB: 60B HEADER 24B ELEMENTS

ADABAS ADARUN LFP=

1. USER : UID (4b) + CID (4b)2. GLOBAL : ADDS5 (8b), HIGH-ORDER = X’11’

TYPES OF IFBs

84


1. FREE SPACE IS CHAINED.

ADABAS 7.1.2

2. GLOBAL IFBs ARE CHAINED (LIFO).

3. USER IFBs ARE CHAINED (OFF U.Q.E.).

4. U.Q.E. CONTAINS ADDRESSES OF LAST 5 IFBs.

5. SCAN FOR IFB TO OVERWRITE (FB OVERWRITES): ???

85


TRANSLATETIME

SCAN FOREXISTING

(5 MOST RECENT)(UQE CHAIN)(G.IFB CHAIN)

SCAN FOR SPACESCAN FOR F.OVWR

86


NO. CMD MEAN CM FLDS COUNT CMDTIME FBLL1 1 1,000 0.00006 8L1 10 1,000 0.00013 80L1 20 1,000 0.00018 160L1 50 1,000 0.00036 400L1 100 1,000 0.00066 800L1 200 1,000 0.00129 1600

TRANSLATE TIME PER X FLDS

IFBS CMD MEANCM /LFP COUNT CMDTIMEL1 1 1,000 0.00006L1 100 1,000 0.00007L1 500 1,000 0.00016L1 1000 1,000 0.00026L1 2000 1,000 0.00047L1 4000 1,000 0.00088L1 8000 1,000 0.00168L1 16000 1,000 0.00362L1 20000 1,000 0.00460

TIME TO SCAN X IFBs

87


a. IF NO G.IFBs MAX LFP

-HIGHWATER < 99%-NO FORMAT OVERWRITES

b. LIMIT G.IFB CHAIN: MOST USED ? ? ?

c. IF LARGE G.IFB CHAIN TOO SMALL < LFP < TO BIG

WATCH CPUDO NOT WATCH FB.TRANSDO NOT WATCH FB.OVWR

88

B. CPU. SOLUTIONS.

(1.5 x HIGH.WATER) < NC < (2 x HIGH.WATER) 8. ADARUN NC.

(1.5 x HIGH.WATER) < NU < (2 x HIGH.WATER) NT <= 99% OF COMMANDS

9. ADARUN NU,NT.

89

B. CPU. SOLUTIONS. 10.DVT CONSTRUCTION.

ELIMINATE BAD DESCRIPTORS.

LARGEST USED NILARGEST ASSO-IOMOST UPDATE CMDS

FILE CANDIDATES:

L3DE: SUMMARIZE(SUBTOT=OFF) COUNT (PF=N9) MIN(NAT-PROGRAM) (PF=A8) MAX(NAT-PROGRAM) (PF=A8) MIN(CID) (PF=H8) MAX(CID) (PF=H8) BY FILE (PF=N3) BY L3-SEQUENCE (PF=A2) WHERE (FILE = 48 OR = 139) AND (CMD = 'L3' OR = 'L6') INTERVAL X HOURS PRINT TO OUTFILE TITLE='L3 DE USE';

90

B. CPU. SOLUTIONS. 10.DVT CONSTRUCTION. KILL BAD DEs.

SXDE: SUMMARIZE(SUBTOT=OFF) COUNT MIN(NAT-PROGRAM) MAX(NAT-PROGRAM) MIN(CID) (PF=H8) MAX(CID) (PF=H8) BY FILE (PF=N3) BY DESCRIPTOR (PF=A2) WHERE (FILE = 48 OR = 139) AND (CMD = 'S1' OR = 'S2' OR = 'S4' OR = 'L9') INTERVAL X HOURS PRINT TO ETWAS TITLE='SX DE USE';

LOGSB

91

B. CPU. SOLUTIONS. 10.DVT CONSTRUCTION. KILL BAD DEs.

BEFORE AFTER CHANGE ------ ------ ------ Number of DE's: 2120 1951 -8% Total DE bytes/recs: 41,248 34,671 -16% Total No. of Records: 231.9M 232.6M 0%Used Data Storage Blocks: 10.3M 10.4M +1% Used Index Blocks: 18.1M 13.4M -26% Cmd time / update cmd: .00110 .00087 +21% Duration / update cmd: .01221 .00696 +43%

RESULTS

92

B. CPU. SOLUTIONS. 11.COMPRESSION/DECOMPRESSION.

! PROGRAMS SHOULD SELECT ONLY THE FIELDS NEEDED !

! AVOID GENERALIZED VIEWS !

A. GENERAL : READS 214 FIELDS OFF FILEB. SPECIFIC: READS 20 FIELDS OFF FILE

DO 100,000 READS WITH BOTH METHODS.

93

B. CPU. SOLUTIONS. 11.COMPRESSION/DECOMPRESSION.

PERCENT SPECIFIC GENERAL CHANGE -------- -------- ------RECORD BUFFER 115b 1718bELAPSED TIME 9.4 MIN 14.7 MIN 56%CPU OF USER ? ? 90%CPU OF ADABAS 89 SEC 152 SEC 71%

RESULTS:

ADABAS SQL: AVOID SELECT *

? FIELDS AT THE END VERSUS THE BEGINNING ?

? FIXED VERSUS COMPRESSED ?

94

B. CPU. SOLUTIONS. 12.BAD CACHE PARMS.

ADARUN CDATAMAXS=1,000,000 CMAXCSPS=1

CDATADSP = 1 – 1,000,000

CACHE SIZE = 197 RABNS

95

B. CPU. SOLUTIONS.

13.ADABAS RELEASE.

14.CPU HARDWARE.

96

C. IO ISSUES. TOOLS.

1. IO / CMD

a. SYNS-60

b. CLOG

c. DDPRINT

2. IO TIMES

a. CLOG

b. CACHE STATS

c. OMEGAMON,SMF, etc

3. BAD PACKS

a. DDPRINT

b. OMEGAMON,SMF, etc

4. BUSIEST FILES

97

C. IO ISSUES. TOOLS. 1. IO / CMD

(SEE B. CPU ISSUES. TOOLS. 3. SYNS-60a. SYNS-60

b. CLOG

IOREP: SUMMARIZE(SUBTOT=NO) COUNT (PF=N5) MEAN(TOTAL-IO) (PF=N1.2) SUM(ASSO-IO) (PF=N5) SUM(DATA-IO) (PF=N5) SUM(WORK-IO) (PF=N3)* BY JOBNAME BY FILE BY CMD* INTERVAL 15 MINUTES* WHERE ETWAS* PRINT TO OUTDD INSIGHT-LINES=300;

98

C. IO ISSUES. TOOLS. 1. IO / CMD

b. CLOG

MEAN SUM SUM SUM CMD TOT ASSO DATA WORK COUNT IO IO IO IO------ ---- ------ ------ ----51,944 0.18 2,732 6,392 0

NOTE: IF ADARUN LFIOP>0, THEN WRITE IO’S ARE NOT INCLUDED

NOTE: IO’S FROM ADABAS CACHE ARE NOT SUBTRACTED OUT( ACTUAL IO WILL BE LESS )

99

C. IO ISSUES. TOOLS. 1. IO / CMD b. DDPRINT

READS WRITESASSO 34656619 19657470DATA 248086098 1791425WORK 4305 27968723PLOG 1522 24974701CLOG 0 0TOTAL 282748544 74392319

A D A B A S EXECUTED 1865226,653 CALLSADAI03 DDWORKR1 4305 READS 27968723 WRITESADAI03 DDDATAR1 248086098 READS 1791425 WRITESADAI01 DDLOG 0 WRITESADAI03 DDPLOGR1 762 READS 12830079 WRITESADAI03 DDPLOGR2 762 READS 12144623 WRITESADAI03 DDASSOR1 34656620 READS 19657471 WRITES

NOTE: CACHE IO SUBTRACTED OUT

NOTE: IF ADARUN LFIOP>0, THEN WRITE IO’S ARE NOT RABNs

100

C. IO ISSUES. TOOLS. 1. IO / CMD b. DDPRINT cont.

VOL-SER HIGH RABN COUNT ADAB59 (ASSO 600828) 6108205 ADAB60 (ASSO 1201668) 3185966 ADAB61 (ASSO 1802508) 29549429 : : : : : : : : ADABDB (DATA 5507695) 13777324 ADABDC (DATA 5758045) 1453608 ADABDD (DATA 6008395) 1TOTAL 419381673

NOTE: CACHE IO NOT SUBTRACTED OUT

NOTE: IO COUNT IS READ AND WRITES

<= VOLSER COUNT

101

C. IO ISSUES. TOOLS. 1. IO / CMD b. DDPRINT cont.

+ ------------ ---ASSO---- ---DATA---- ---WORK--- ++ CACHE WRITES 103,043,539 83,878,641 0 ++ READ EXCPS 3,380,279 68,183,677 0 ++ CACHE READS 99,521,064 15,669,051 0 ++ TOTAL READS 102,901,343 83,852,728 0 ++ EFFICIENCY 96.7 18.6 N/A ++-------------------------------------------------+

NOTE: ASSO.READS + ASSO.WRTS + DATA.READS + DATA.WRTS + CACHE.ASSO.READS + CACHE.DATA.READS= VOLSER.COUNT

EXAMPLEEXAMPLE34656619 + 19657470 + 248086098 + 1791425 + 99,521,064 + 15,669,051= 419381673 (+/- COUPLA IO’S)

102

C. IO ISSUES. TOOLS. 2. IO TIMES a. CLOG

IOTIME: SUMMARIZE(SUBTOT=NO) COUNT (PF=N5) MEAN(DURATION) (PF=N1.5) MEAN(CMD-TIME) (PF=N1.5) BY CMD BY TOTAL-IO (PF=N1 HD='IO') WHERE ECBS = 1 AND TOTAL-IO LE 1 AND CMD(1) = 'L'* AND FILE = 111

C MEAN MEAN M CMD DURAT CMD D IO COUNT SECS TIME-- -- ------ ------- -------L3 0 4,889 0.00018 0.00005L3 1 276 0.00150 0.00022L9 0 216 0.00022 0.00012L9 1 3 0.01303 0.00023

103

C. IO ISSUES. TOOLS. 2. IO TIMES b. CACHE

+-------------------------------------------------------++ 00022 DATA SPACE, ASSO, RABNS 14,986,838 THRU 15,008,837+-------------------------------------------------------++ ALLOCATED, LA=01:34:16 ++-------------------------------------------------------++ 3,555,713 CACHE WRITES + 5,368 BLKS IN CACHE ++ 416,188 READ EXCPS + 200,007,680 CACHE SIZE ++ 3,120,907 CACHE READS + 48,350 BLKS/CACHE ++ 3,537,095 TOTAL READS + 88,016 RABN TAB SIZE ++ 88.2 DSP EFFICIENCY+ 36,446,432 MAX DSP USED ++ 0.526364 MAX NIOT (SEC)+ 1.221436 MAX EXCPT(SEC)++ 0.000003 MIN NIOT (SEC)+ 0.001940 MIN EXCPT(SEC)++ 0.000077 AVE NIOT (SEC)+ 0.028199 AVE EXCPT(SEC)++-------------------------------------------------------+

104

C. IO ISSUES. TOOLS. 2. IO TIMES

+========================== DASD Device Activity ========|Volume SMF Dev I/O -------- Time in Milliseconds|Serial id # LCU Rate Total = IOSQ+Pend+Conn+Disc |------ ---- ---- --- ---- ----- ---- ---- ---- ---- -|ADABC2 CNT1 085F 00E 6.2 1 0 0 1 0

c. OMEGAMON,EPILOG,MAINVIEW,ASTEX,etc

105

C. IO ISSUES. TOOLS. 3. BAD PACKS

+------------------------------------------------| STARTED TASK = DB999NUC JES NUMBER = 15859| FROM 00:14 ON 10/04/01 TO 00:14 ON 10/05/01+------------------------------------------------|WAIT_REASON_____________TIME_____%_||USING CPU 9:35 H 39.9||ECB WAIT (W/ STIMER) 3:28 H 14.4||DISK ADAB58 1033 ACT 1:28 H 6.1||STC ELAPSED TIME 23:59 H+------------------------------------

a. DDPRINT VOLSER COUNTS (SEE EARLIER)

b. OMEGAMON,EPILOG,MAINVIEW,ASTEX,etc

106

C. IO ISSUES. TOOLS. 4. BUSIEST FILES.

BIGFILE: SUMMARIZE(SUBTOT=NO) COUNT (PF=N9) SUM(ASSO-IO) (PF=N9) SUM(DATA-IO) (PF=N9) SUM(DURATION) (PF=N6.1) SUM(CMD-TIME) (PF=N6.1) BY FILE (PF=N3)

FILE COUNT ASSOIO DATAIO DURATION CMDTIME 39 120,421,719 6,723,425 16,394,000 145,645.6 8,994.4 86 156,184,181 3,855,282 9,389,882 117,031.2 11,787.5 196 69,001,326 2,627,615 7,888,037 88,886.2 6,548.5 93 41,769,213 3,244,980 8,372,643 88,467.9 4,491.9 108 84,563,613 3,973,571 8,485,772 79,516.9 6,737.5 90 92,297,406 4,928,471 6,343,287 71,601.1 5,684.3 95 43,576,891 1,834,556 1,878,552 35,708.2 3,020.4

SORT: BY COUNT, BY DURATION, BY ASSO/DATA IO, BY CMDTIME

107

C. IO ISSUES. SOLUTIONS.

1. FORWARD INDEX COMPRESSION. 2. REMOVAL OF UNNEEDED DESCRIPTORS. 3. CHOICE OF DASD. 4. ADARUN ASYTVS=. 5. FILE PLACEMENT. 6. BLOCKSIZE. 7. REORDERs. 8. REUSE ISN/DS. 9. ASSO/DATA PADDING.10.WORK 3 CACHE.11.WORK 2 CACHE.12.ADABAS CACHING.

108


SEE NEW WITH ADABAS 71. FORWARD INDEX COMPRESSION.

2. REMOVAL OF UNNEEDED DESCRIPTORS.SEE B.CPU SOLUTIONS: 10.DVT CONSTRUCTION.

3. CHOICE OF DASD.

4. ADARUN ASYTVS=.SEE NEW WITH ADABAS 7. LEAVE AS DEFAULT (=YES)

BEYOND THE SCOPE OF THIS PRESENTATION.

109


5. FILE PLACEMENT.

SEE DDPRINT VOLSER COUNTS (MINUS CACHE)SEE SYSTEM REPORTS (OMEGAMON,RMF,ETC)SEE BUSIEST FILE REPORTSSEE ADAREP FOR WHICH FILES PER VOLSER

SEE WORST

SOLUTION

MOVE THEM, SPREAD THEM

110

C. IO ISSUES. SOLUTIONS. 6. BLOCKSIZE.

1. MAY IMPROVE PERFORMANCE ON READS.2. ALLOW LARGER COMPRESSED RECORD SIZES.3. FURTHER AVOIDANCE OF WORK WRAPAROUNDS.

ADVANTAGES

? ? ?DISADVANTAGES

ADAORD RESTRUCTUREDB, ADAFRM ASSO,DATA,etc,ADADEFADAORD STORE

PROCESS

111

C. IO ISSUES. SOLUTIONS. 6. BLOCKSIZE. ADARUN PARM CHANGES

DEVICE = 8391 (everywhere)

LBP = 2 X OLD.LBP

WORK: EITHER DOUBLE WORKSIZE or LP = .42 X OLD.LP LWKP2 = .42 X OLD.LWKP2

DUALPLD = 8391DUALPLS = 60 X PLOG.CYLINDERS

CASSOMAXS = 2 X OLD.CASSOMAXSCDATAMAXS = 2 X OLD.CDATAMAXS

112


7. REORDERs. : WHAT, WHEN ?

8. REUSE ISN/DS: USUALLY ON, UNLESS

9. ASSO/DATA PADDING: MIN. FOR BIG FILES

113

C. IO ISSUES. SOLUTIONS. 10.WORK 3 CACHE.

1. LOTSA WORK READS (?)2. JUST TRY IT ( 1 %).3. Sx ANALYSIS

WHEN TO: SX: SUMMARIZE COUNT (PF=N7) MEAN(ISNQ) (PF=N4.1) MEAN(WORK-IO) (PF=N2.1) MAX(WORK-IO) (PF=N3) BY ISNQ (PF=N5 BY 50) WHERE CMD(1) = 'S'

ADARUN CWORK3FAC= %ADARUN CWORKSTORAGE=DATASPACE

MEAN MEAN MAX ISN CMD ISN WORK WORKQUANT COUNT QUANT IO IO 0 247,283 2.7 0.0 0 50 101 85.8 1.0 1 100 17 134.6 0.9 1 : : : : : : : :7,600 3 7,635.0 3.0 3===== 247,467 3.0 0.0 3

114

C. IO ISSUES. SOLUTIONS. 10.WORK 3 CACHE.

DD- I DEV I NMBR OF I NMBR OF NAMES I TYPE I CYLS I BLOCKS WORKR1 I 3390 I 500 I 67491

1.DETERMINE W3

BLOCKS

ADAREP:

LP=50000MINUS ADARUN: LWKP2=14000MINUS ADARUN: 3491 W3 SIZE

= 2.WORST CASE NO. OF LISTS

TBI -POOL LI = 100000 29080 ( 29 %)DDPRINT:

(52 + 4 X NSISN)ONE ELEMENT=

= 352b (NSISN=75)

DIVIDE TBI BY ELEMENT SIZE 82 NO. OF LISTS

ADARUN CWORK3FAC= 82/3491 = 3 % OR LESS

ASSUME 1 WORK BLOCK PER LIST

115


10.WORK 2 CACHE (ADARUN CWORK2FAC=):

a. LESS THAN 50%

b. TRY IT

MONITOR BOTH WORK 3 AND 2 CACHE SPACE VIA CACHE STATS

RE-EVALUATE CWORK3FAC / CWORK2FAC IF A CHANGE IN:

1. WORK BLOCKSIZE 2. ADARUN LWKP2= 3. WORKSIZE (JUST CWORK3FAC) 4. ADARUN LP= (JUST CWORK3FAC)

116

C. IO ISSUES. SOLUTIONS. 12.ADABAS CACHING.

THE RESULTS:

SAVE TOTAL.CMDS TOTAL.RDS CACHE.RDS PCT% CBE PBE SAVDUR1,509,095,970 297,266,628 123,654,424 29.4 4.3 3.0 0.00162,438,702,209 198,852,925 82,473,849 29.3 6.0 4.2 0.00083,440,990,996 285,295,358 120,069,839 29.6 2.4 1.6 0.00072,729,211,969 313,548,268 130,706,933 29.4 6.8 4.7 0.0012

CBE: CURRENT ADABAS BUFFER EFFICIENCY PBE: PROJECTED NON-CACHE BUFFER EFFICIENCY

SAVDUR: AVERAGE DURATION SAVED PER COMMAND (RESULT=0.00085

SAVDUR= SUM OF ALL (CACH.READ * (AVE.EXCP - AVE.NIOT)) DIVIDED BY TOTAL.CMDS

117

C. IO ISSUES. SOLUTIONS. 12.ADABAS CACHING. ADARUNS

CACHE = YESCMAXCSPS = 1 * USELESS SINCE V6.2CASSOMAXS = 100-200M * WATCH ZEROES!CDATAMAXS = 100-200M * WATCH ZEROES! CDISPSTAT = NO

XA vs DSP vs EXT

ADARUNS:

NOTE ADABAS REGION: 4b PER RABN IN RANGEEg. CDATADSP=2000000-3000000

1,000,000 RABNS4M IN ADABAS

118

C. IO ISSUES. SOLUTIONS. 12.ADABAS CACHING. TYPE OF ?

1.CACHE RABN RANGES, CASSODSP=from-to

2.CAN START ANYTIME

3.REVIEW IF FILE MOVES.

4.CACHE MAINTAINED FIFO.

5.CAN CACHE JUST UI, AC, NI, FCBs, FDTs, DSST

1.CACHE FILE, CLASS, ASSO/DATA/BOTH

2.MUST START AT NUC INIT.

3.ADAJUSTS IF FILE MOVES. 4.CACHE MAINTAINED BY CLASS BY FIFO

5.CACHE ALL FILE ASSO. CANNOT CACHE FCB,FDT, SEPARATE UI,AC,NI.

GLOBAL CACHING versus FILE CACHING

119

C. IO ISSUES. SOLUTIONS. 12.ADABAS CACHING.

WHO / WHAT TO CACHE PICK TOP FILES BY:

FNR COUNT ASSOIO DATAIO DURATION CMDTM 39 120421719 6723425 16394000 145645.6 8994.4 86 156184181 3855282 9389882 117031.2 11787.5196 69001326 2627615 7888037 88886.2 6548.5 93 41769213 3244980 8372643 88467.9 4491.9 : : : : : : : : : : :

CMDTIMEDURATIONASSO OR DATA IOCOUNTCRITICALNESS

1. CACHE FCB-FDT,DSST.

2. CACHE TOP 10 TO 30 FILES UI / AC.

3. CACHE TOP 5 TO 15 FILES NI. ( WATCH NI USED RABNS !)

4. CACHE TOP 1 TO 2 FILES DS WHERE USED RABNS < 2M

120

C. IO ISSUES. SOLUTIONS. 12.ADABAS CACHING. HOW TO

*FNR TYPE GROWTHFCBDSST39 AC 100%39 UI 100%39 NI 0%93 AC 100%93 UI 100%93 NI 0%93 DS 0%86 AC 100%86 UI 100%49 UI 100%49 AC 100%: :

CACHE TABLE1. CREATE CACHE TABLE OF

REQUESTS.

2. AT DATABASE START RUN CACHE JOB:

STEP 1 – ADAREP

STEP 2 – CACHE PROGRAM

STEP 3 – ISSUE CACHE COMMANDS

1.READ TABLE.2.READ ADAREP FOR FILE, TYPE, RABN RANGES.3.DETERMINE USED RABN RANGE.4.ADJUST TO GROWTH.5.BUILD CACHE OPERATOR COMMANDS.

CACHE PROGRAM

121

C. IO ISSUES. SOLUTIONS. 12.ADABAS CACHING. HOW TO Count of Blocksize

Fnr/Type FromRabn ToRabn Rabns * RabnCntFCB... ASSO ______31 ____1305 ____1275 5273400DSST.. ASSO ____1306 ____1850 _____545 2254120DSST.. ASSO ____8765 ____8825 ______61 252296DSST.. ASSO __291973 __292153 _____181 748616.14 AC ASSO __345527 __346253 _____727 3006872.14 NI ASSO _2312178 _2318177 ____6000 24816000.14 UI ASSO ____8595 ____8674 ______80 330880.21 AC ASSO __469025 __469242 _____218 901648.21 NI ASSO __336926 __337925 ____1000 4136000.21 UI ASSO ___13004 ___13033 ______30 124080: : : : : : : :

TOTAL ASSO RABNS REQUESTED TO BE CACHED: 478328TOTAL DATA RABNS REQUESTED TO BE CACHED: 77987

XA MEMORY REQUIREMENTS FOR ASSO RABNINDEX ARE: 1913312XA MEMORY REQUIREMENTS FOR DATA RABNINDEX ARE: 311948

LIST OF CACHE COMMANDS TO BE ISSUED:/F DB044NUC,CASSODSP=31-1305 FCB/F DB044NUC,CASSODSP=1306-1850 DSST

122

A D A B A S DYNAMIC CACHING SESSION S U M M A R Y 6.8 CURRENT ADABAS BUFFER EFFICIENCY 4.7 PROJECTED NON-CACHE BUFFER EFFICIENCY 51 ACTIVE RABN RANGES 51 RABN RANGES DEFINED------------ -----ASSO---- -----DATA---- -----WORK----CACHE WRITES 125,209,233 29,452,553 0 READ EXCPS 7,639,638 15,928,509 0 CACHE READS 117,208,374 13,498,559 0 TOTAL READS 124,848,012 29,427,068 0 EFFICIENCY 93.8 45.8 N/A------------------------------------------------------MAX CACHE 224,640K 195,320K

00007 DATA SPACE, ASSO, RABNS 13,550 THRU 71,653 ALLOCATED, LA=01:33:18 9,196,498 CACHE WRITES + 9,387 BLKS IN CACHE 809,794 READ EXCPS + 230,031,360 CACHE SIZE 8,383,500 CACHE READS + 55,610 BLKS/CACHE 9,193,294 TOTAL READS + 232,432 RABN TAB SIZE 91.1 DSP EFFICIENCY+ 53,160,008 MAX DSP USED 0.395320 MAX NIOT (SEC)+ 77.639942 MAX EXCPT(SEC) 0.000004 MIN NIOT (SEC)+ 0.001938 MIN EXCPT(SEC) 0.000052 AVE NIOT (SEC)+ 0.024895 AVE EXCPT(SEC)

C. IO SOLUTIONS

12. ADABAS CACHING

TRACKING

DDPRINT

123

C. IO ISSUES. SOLUTIONS. 12.ADABAS CACHING. TRACKING.

SESSION PCT PCTDATE HOURS STRUC CACHREADS CACH EFF BLOCKS CACHWRTS20010520 670.3 108NI 14362597 5.3 90.1 55224 1593957520010617 310.6 108NI 5831633 5.3 89.2 55751 653299120010630 358.8 108NI 6920049 5.6 90.2 56042 767294120010715 286.6 108NI 6097142 5.9 90.7 56274 671916920010727 215.3 108NI 4824751 5.9 89.5 56700 539194620010805 307.9 108NI 6360703 5.3 89.7 56730 708755920010827 296.5 108NI 6759524 6.1 90.6 57274 745800420010909 334.0 108NI 8383500 6.4 91.1 58104 9196498

READEXCPS AVE.NIOT AVE.EXCP 1572347 0.000050 0.019529 699255 0.000051 0.019611 751012 0.000048 0.018708 620154 0.000046 0.023230 565448 0.000054 0.022807 724404 0.000052 0.020300 695901 0.000051 0.021086 809794 0.000052 0.024895

. . .

124

THE END

1 ADABAS 7 PERFORMANCE CHECKLIST October, 2001 Jim Poole.

Documents

lfiop access

lbp access

index space

backup lbp

cpu improvement

faster utilities

cpu cmdtime differenceschart10

overall database space