DB2 UDB for z/OS Version 8 Performance Topics - IBM Redbooks

More documents

Recommendations

Info

Star Schema Figure 3-15 Star schema The attributes of a star schema database are: ► Large fact table: Like sales tables containing transactions that can be in the order of hundreds of millions, or billions of data rows. ► Highly normalized design with dimensions (or snowflakes), tables to avoid maintaining redundant descriptive data in the central fact table. ► Relatively small dimensions: Dimensions can be denormalized to one table (without the tables identified as “S” in Figure 3-15) or normalized in related tables in the form of a snowflake. ► Sparse “Hyper Cube”: Caused by high correlation among dimensions, leading to a sparse nature of data in the fact table; for example, product sales are dependent on the climate, therefore, the sale of shorts is more likely in a state where the people enjoy hot weather. ► Fact table is dependent on the dimension tables. Star schema query Star schema query is normally a star join with the following characteristics: ► Equi-join predicates between the fact and dimension tables ► Local predicates on the dimension tables ► Large number of tables in the query For a star join query, DB2 uses a special join type called a star join if the following conditions are true: ► The tables meet the conditions that are specified in the section “Star join (JOIN_TYPE=’S’)” of the DB2 UDB for z/OS Version 8 Administration Guide, SC18-7413. Unlike the steps in the other join methods (nested loop join, merge scan join, and hybrid join) in which only two tables are joined in each step, a step in the star join method can involve three or more tables. Dimension tables are joined to the fact table via a multi-column index that is defined on the fact table. Therefore, having a well-defined, multi-column index on the fact table is critical for efficient star join processing. ► The STARJOIN system parameter is set to ENABLE, and the number of tables in the query block is greater than or equal to the minimum number that is specified in the SJTABLES system parameter. Note: The default for star join processing is DISABLE. ► Another system parameter is the maximum pool size (MB) for star join SJMXPOOL. 54 DB2 UDB for z/OS Version 8 Performance Topics S D Fact table D F D D S S snowflake Dimension table S S
Star join technical issues The technical issues surrounding the optimization of decision support and data warehousing queries against a star schema model can be broken down into bind-time and run-time issues. The first consideration generally is the sheer size of the fact table, and also the large number of tables that can be represented in the star schema. ► Bind-time issues For non-star join queries, the optimizer uses the pair-wise join method to determine the join permutation order. However, the number of join permutations grows exponentially with the number of tables being joined. This results in an increase in bind time, because of the time required to evaluate the possible join permutations. Star join does not do pair-wise join. It joins several unrelated tables via a (simulated) cartesian join to the fact table. ► Run-time issues Star join queries are also challenging at run-time. DB2 uses either cartesian joins of dimension tables (not based on join predicates, the result is all possible combinations of values in both tables) or pair-wise joins (based on join predicates). The result is all possible combinations of values in both tables. The pair-wise joins have worked quite effectively in the OLTP world. However, in the case of star join, since the only table directly related to other tables is the fact table, the fact table is most likely chosen in the pair-wise join, with subsequent dimension tables joined based on cost. Even though the intersection of all dimensions with the fact table can produce a small result, the predicates supplied to a single dimension table are typically insufficient to reduce the enormous number of fact table rows. For example, a single dimension join to the fact table may find: – 100+ million sales transactions in the month of December – 10+ million sales in San Jose stores – 10+ million sales of Jeans, but – Only thousands of rows that match all three criteria Hence there is no one single dimension table which could be paired with the fact table as the first join pair to produce a manageable result set. With these difficulties in mind, the criteria for star join can be outlined as follows: ► “Selectively” join from the outside-in: Purely doing a cartesian join of all dimension tables before accessing the fact table may not be efficient if the dimension tables do not have filtering predicates applied, or there is no available index on the fact table to support all dimensions. The optimizer should be able to determine which dimension tables should be accessed before the fact table to provide the greatest level of filtering of fact table rows. For this reason, outside-in processing is also called the filtering phase. DB2 V8 enhances the algorithms to do a better job at selecting which tables to join during outside-in processing. ► Efficient “Cartesian join” from the outside-in: A physical cartesian join generates a large number of resultant rows based on the cross product of the unrelated dimensions. A more efficient cartesian type process is required as the number and size of the dimension tables increase, to avoid an exponential growth in storage requirements. A key feedback technique is useful for making cartesian joins efficient. ► Efficient “join back”, inside-out: The join back to dimension tables that are accessed after the fact table must also be efficient. Non-indexed or materialized dimensions present a challenge for excessive sort Chapter 3. SQL performance 55
Page 1:
ibm.com/redbooks Front cover DB2 UD
Page 4 and 5:
Note: Before using this information
Page 6 and 7:
2.5.3 Increased number of deferred
Page 8 and 9:
4.5.2 Conclusion . . . . . . . . .
Page 10 and 11:
7.4.1 ODBC Unicode support . . . .
Page 12 and 13:
x DB2 UDB for z/OS Version 8 Perfor
Page 14 and 15:
3-48 Consistency query - 31 New . .
Page 16 and 17:
xiv DB2 UDB for z/OS Version 8 Perf
Page 18 and 19:
8-5 CF Request Batching - DB2 CPU .
Page 20 and 21:
xviii DB2 UDB for z/OS Version 8 Pe
Page 22 and 23:
Trademarks The following terms are
Page 24 and 25:
Added information The following APA
Page 26 and 27:
► Added reference to June 15, 200
Page 28 and 29:
Michael Parbs is a DB2 Specialist w
Page 30 and 31:
Hugh Smith Jim Teng John Tobler Yoi
Page 32 and 33:
xxx DB2 UDB for z/OS Version 8 Perf
Page 34 and 35:
1.1 Overview 1.1.1 Architecture IBM
Page 36 and 37: - Column data type changes can be d
Page 38 and 39: 1.1.3 Data warehouse 1.1.4 Performa
Page 40 and 41: Volatile (or cluster) tables, used
Page 42 and 43: In this chapter we anticipate the a
Page 44 and 45: 2.1.3 I/O time 2.1.4 Virtual storag
Page 46 and 47: ► The performance benchmarks for
Page 48 and 49: application that can have great pot
Page 50 and 51: Because you can significantly incre
Page 52 and 53: The possible increase in thread foo
Page 54 and 55: 2.5.9 Unicode parser Important: Con
Page 56 and 57: You can keep the DBRMs in EBCDIC fo
Page 58 and 59: Macro Parameter Old value New value
Page 60 and 61: 28 DB2 UDB for z/OS Version 8 Perfo
Page 62 and 63: ► Parallel sort enhancement Paral
Page 64 and 65: Characteristics of the host variabl
Page 66 and 67: 3.1.4 Performance Update 10 rows us
Page 68 and 69: ► Complexity of the fetch. The fi
Page 70 and 71: ► More rows per one SQL statement
Page 72 and 73: 3.2.1 Creating MQTs MQTs compared t
Page 74 and 75: Special register CURRENT REFRESH AG
Page 76 and 77: the materialized query table. If it
Page 78 and 79: Figure 3-9 MQTs and short running q
Page 80 and 81: ► The elapsed time is 22 times sm
Page 82 and 83: Figure 3-13 Regrouping on MQT resul
Page 84 and 85: EVE.EXT = 'ADD BILL' AND EVE.TXC =
Page 88 and 89: merge joins and work file usage. DB
Page 90 and 91: Star join access path without a spa
Page 92 and 93: 3.3.4 Performance Sparse index for
Page 94 and 95: V8 Star Join Access Path Figure 3-2
Page 96 and 97: 3.3.5 Conclusion Star join workload
Page 98 and 99: . SELECT * FROM PAOLO.TORDER WHERE
Page 100 and 101: But not all mismatched numeric type
Page 102 and 103: V8 shows that PL is joined with CV
Page 104 and 105: Query performance problem Lack of m
Page 106 and 107: 3.5.1 Performance SORTDEVT SYSDA IN
Page 108 and 109: 3.5.2 Conclusions Figure 3-31 Acces
Page 110 and 111: Mismatched data types SELECT COUNT(
Page 112 and 113: 3.7.1 Performance In V8 the paralle
Page 114 and 115: 3.7.2 Conclusions Example 3-9 Multi
Page 116 and 117: 3.8.2 Performance We show three exa
Page 118 and 119: 3.8.3 Conclusion The measurements s
Page 120 and 121: 100 Inserts activated trigger 100 t
Page 122 and 123: 3.10.2 Conclusion 100 Inserts activ
Page 124 and 125: stores the selected columns from ea
Page 126 and 127: While there is no optimistic lockin
Page 128 and 129: FETCH FROM Figure 3-45 FETCH syntax
Page 130 and 131: ► Measurement data collected with
Page 132 and 133: application has decidedly higher Cl
Page 134 and 135: Static updatable cursor You can see
Page 136 and 137:
Multi-row FETCH with a static curso
Page 138 and 139:
3.12.5 Conclusion Multi-row FETCH a
Page 140 and 141:
While, for or SELECT ... FROM ...
Page 142 and 143:
3.14.2 Conclusion Percent reduction
Page 144 and 145:
3.15.1 Installation commands. Of co
Page 146 and 147:
Figure 3-47 Explain of the old cons
Page 148 and 149:
in ms = 1 and the CPU cost is 16 se
Page 150 and 151:
Figure 3-50 Suggested RUNSTATS stat
Page 152 and 153:
Looking at the explanation, we can
Page 154 and 155:
The Plug-In Options screen of Stati
Page 156 and 157:
Before the introduction of MQTs, DB
Page 158 and 159:
126 DB2 UDB for z/OS Version 8 Perf
Page 160 and 161:
► Larger VSAM CI sizes: DB2 V8 in
Page 162 and 163:
4.1.1 Performance support caused al
Page 164 and 165:
Regression is about 5% better than
Page 166 and 167:
The results indicate no significant
Page 168 and 169:
Important: 4.1.2 Conclusion If you
Page 170 and 171:
However, the combined impact of oth
Page 172 and 173:
Frees up space below the bar for ot
Page 174 and 175:
from the hiperpools, they were not
Page 176 and 177:
So, we advise you to not overalloca
Page 178 and 179:
See 4.3, “Monitoring DBM1 storage
Page 180 and 181:
Table 4-4 System storage configurat
Page 182 and 183:
Stack 11356 21033 85% 8648 19799 12
Page 184 and 185:
KB 600 500 400 300 200 100 0 Figure
Page 186 and 187:
For the same buffer pool size and o
Page 188 and 189:
4.2.3 Conclusion Nondistributed per
Page 190 and 191:
Important: With PTFs UK04394 (DB2 V
Page 192 and 193:
- DATASPACE BP CONTROL BLOCKS = 0 M
Page 194 and 195:
The EXTENDED REGION SIZE (MAX) is t
Page 196 and 197:
RDS OP POOL (MB) N/A RID POOL (MB)
Page 198 and 199:
As you can now appreciate, it is im
Page 200 and 201:
Megabytes A V7 Customer DBM1 Thread
Page 202 and 203:
4.5.1 Performance The most buffer p
Page 204 and 205:
4.6.1 Performance maintained in a 2
Page 206 and 207:
4.6.2 Conclusion We can conclude th
Page 208 and 209:
service for Unicode conversions, an
Page 210 and 211:
overcome this, when DB2 Connect is
Page 212 and 213:
INSERT HV1 1200 HV2 1200 SELECT HV1
Page 214 and 215:
AIX C-program Conversion Routines P
Page 216 and 217:
4.7.1 Performance The improvements
Page 218 and 219:
In October 2003, a new enhancement
Page 220 and 221:
UseServerEncoding data source prope
Page 222 and 223:
4.8.1 Performance 4.8.2 Conclusion
Page 224 and 225:
Multilevel security with row-level
Page 226 and 227:
► The security level that defines
Page 228 and 229:
DB2 must call RACF for this dominan
Page 230 and 231:
4.9.2 Conclusion The CPU increase c
Page 232 and 233:
4.10.1 Performance VSAM does not kn
Page 234 and 235:
4.10.2 Conclusion 4.10.3 Striping A
Page 236 and 237:
5 4 3 2 1 0 9109TLLB 4 Parallel DB2
Page 238 and 239:
--------------------------- -------
Page 240 and 241:
This message is completed with the
Page 242 and 243:
This enhancement is introduced by P
Page 244 and 245:
This record is activated by: -START
Page 246 and 247:
an online transaction with very sim
Page 248 and 249:
Page 250 and 251:
5.1 System level point-in-time reco
Page 252 and 253:
BACKUP SYSTEM During backup DB2 rec
Page 254 and 255:
Figure 5-1 FRBACKUP PREPARE elapsed
Page 256 and 257:
Figure 5-4 BACKUP SYSTEM FULL measu
Page 258 and 259:
5.1.2 Conclusion The BACKUP SYSTEM
Page 260 and 261:
Figure 5-7 shows the sliding scale
Page 262 and 263:
Maybe you have implemented a space
Page 264 and 265:
For dynamically prepared queries, D
Page 266 and 267:
► CHAR(8) to VARCHAR(8) ► CHAR(
Page 268 and 269:
Before ALTER INDEX add column, a ta
Page 270 and 271:
30 25 20 15 10 Figure 5-12 FETCH CP
Page 272 and 273:
Percent of index pages Avg. Index L
Page 274 and 275:
300 250 200 time in seconds 150 100
Page 276 and 277:
REORG REBALANCE We executed and mea
Page 278 and 279:
5.4 Volatile table 5.4.1 Conclusion
Page 280 and 281:
If explicit clustering for a table
Page 282 and 283:
Load table of 20M rows, 10 partitio
Page 284 and 285:
Times in seconds 20 18 16 14 12 10
Page 286 and 287:
RUNSTATS INDEX PARTITION of a DPSI
Page 288 and 289:
Note that when the qualified partit
Page 290 and 291:
► “NO” - Values in such colum
Page 292 and 293:
5.6.10 Support for more than 245 se
Page 294 and 295:
6.1 Online CHECK INDEX Online CHECK
Page 296 and 297:
SORTOUT DD UNIT=SYSDA,SPACE=(CYL,(5
Page 298 and 299:
- SORT 6 indexes in parallel - CHEC
Page 300 and 301:
make DB2 take the maximum paralleli
Page 302 and 303:
6.3 LOAD and UNLOAD delimited input
Page 304 and 305:
6.3.2 Conclusion We measured the CP
Page 306 and 307:
Options for cardinality and distrib
Page 308 and 309:
► Number of column groups: 1 ►
Page 310 and 311:
cases is the total time of RUNSTATS
Page 312 and 313:
6.4.5 Conclusion Statistics Advisor
Page 314 and 315:
goals for end-user services. We sho
Page 316 and 317:
alanced across all members. With th
Page 318 and 319:
method. Prior to DB2 V8 those array
Page 320 and 321:
when compared to DB2 V7. At the ser
Page 322 and 323:
7.2.2 Performance - Distributed cli
Page 324 and 325:
Recommendation If you currently do
Page 326 and 327:
► The DB2 Universal JDBC type 4 D
Page 328 and 329:
WebSphere z/OS - DB2 z/OS V8 - Serv
Page 330 and 331:
DB2 Universal Type 4 Driver - JDBC
Page 332 and 333:
Use DB2 Connect when a high amount
Page 334 and 335:
JCC Type 4 Direct 4:3 Concentration
Page 336 and 337:
The SYSIBM.INDOUBT table and the pa
Page 338 and 339:
Batch updates A batched update is a
Page 340 and 341:
Example In DB2 V8 SQL/XML publishin
Page 342 and 343:
INSERT INTO PUB_TAB(DOC) SELECT XML
Page 344 and 345:
7.5.3 Conclusion The new XML publis
Page 346 and 347:
A second role of the managed server
Page 348 and 349:
7.6.3 Performance The following inf
Page 350 and 351:
Page 352 and 353:
Impact on coupling facility: You do
Page 354 and 355:
Table 8-1 shows extracts from DB2 P
Page 356 and 357:
and secondary GBP structures have t
Page 358 and 359:
DB2 member already has the table sp
Page 360 and 361:
8.2.1 Performance We ran a number o
Page 362 and 363:
esolve any contention. This extra a
Page 364 and 365:
There is also no need to use RELEAS
Page 366 and 367:
Table 8-12 DB2 I/O CI Limit Removal
Page 368 and 369:
8.5 Miscellaneous items In this sec
Page 370 and 371:
DB2 V8 improves Restart Light. If i
Page 372 and 373:
Page 374 and 375:
9.1 Unicode catalog DB2 for z/OS su
Page 376 and 377:
that all applications are converted
Page 378 and 379:
- CLASST: Specifies the threshold a
Page 380 and 381:
From these measurements we observe
Page 382 and 383:
Class 1 CPU time (MSEC) 200 150 100
Page 384 and 385:
dynamic statement cache. If activat
Page 386 and 387:
that everything is ready. This step
Page 388 and 389:
Table Number of rows SYSDATABASE 75
Page 390 and 391:
In Figure 9-7 we describe the perfo
Page 392 and 393:
Old - Query 5 New - Query 5 CPU (se
Page 394 and 395:
Query 35 36 Old DSNTESQ CPU 1 hr 1
Page 396 and 397:
10.1 DB2 Estimator DB2 Estimator V8
Page 398 and 399:
- REAL AND AUXILIARY STORAGE This i
Page 400 and 401:
Another use is during back out of a
Page 402 and 403:
In DB2, the IBM Data Encryption too
Page 404 and 405:
The utilities ran on a G7 Turbo wit
Page 406 and 407:
time Figure 10-1 DB2 HPU vs. UNLOAD
Page 408 and 409:
DSNTIAUL with multi-row fetch The r
Page 410 and 411:
A.1 Maintenance for DB2 V8 The APAR
Page 412 and 413:
APAR # Area Text PTF and Notes PQ88
Page 414 and 415:
Page 416 and 417:
APAR # Area Text PTF and Notes PK21
Page 418 and 419:
Page 420 and 421:
Page 422 and 423:
Page 424 and 425:
Page 426 and 427:
B.1 DB2 PE Storage Statistics trace
Page 428 and 429:
Storage heading Description 24 BIT
Page 430 and 431:
Page 432 and 433:
C.1 DB2 SQL PA additional reports N
Page 434 and 435:
|is the last matching predicate. Ap
Page 436 and 437:
|ANL6052I *** NOTE: | |This index i
Page 438 and 439:
PREDICATE ANALYSIS ----------------
Page 440 and 441:
D.1 DB2 EXPLAIN EXPLAIN describes t
Page 442 and 443:
Table D-2 PLAN_TABLE contents and b
Page 444 and 445:
Column Type Content SORTC_ORDER BY
Page 446 and 447:
Column Type Content OPTHINT CHAR(8)
Page 448 and 449:
D.1.3 DSN_FUNCTION_TABLE You can us
Page 450 and 451:
Example: D-1 Creating DSN_STATEMENT
Page 452 and 453:
Column Type Content BIND_QUALIFIER
Page 454 and 455:
Page 456 and 457:
ICF integrated catalog facility ICF
Page 458 and 459:
► DB2 UDB for z/OS V8: Through th
Page 460 and 461:
Page 462 and 463:
character set 174 CHECK DATA 262 CH
Page 464 and 465:
Equi-join predicates 54, 402 Equiva
Page 466 and 467:
MINSTOR 145-146 Mismatched data typ
Page 468 and 469:
PQ84160 378, 387 PQ85764 387 PQ8584
Page 470 and 471:
Sort tree nodes 144 SORTDATA 268 SO
Page 472 and 473:
306 Universal Driver for SQLJ and J
Page 474 and 475:
Page 476:
DB2 UDB for z/OS Version 8 Performa
show all

DB2 UDB for z/OS Version 8 Performance Topics - IBM Redbooks

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?