Starter’s Guide to Db2 for z/OS

Data Sharing Monitoring and

Tuning

John Campbell
Session code:
Distinguished Engineer
Db2 for z/OS Development • Platform: Db2 for z/OS

IBM Data and AI

Agenda
• Global locking
• Db2 LOCK1 structure
• Global contention monitoring and tuning

• Group buffer pools

• GBP reads
• GBP writes
• GBP castout
• GBP monitoring and tuning

2
Db2 LOCK1 structure Db2 Lock Structure

Lock Hash Table

• What is it used for? Exclusive Shared Lock Status
1 00 1 0 0 0 0 0 0
• Fast inter-system global lock conflict detection 2 02 0 0 0 0 0 0 0

• Optimisation for fast grant of global lock where no contention

m 00 1 1 0 0 0 0 0
• Fast inter-system page latch conflict detection for cases
where sub-page concurrency is allowed q 02 0 0 0 0 0 0 0

• Row level locking, space map pages, index leaf pages n 02 0 0 0 0 0 0 0

DB2B
• Inter-system read/write interest tracking for Db2 objects IRLM Record List Table (Modified Lock List)
• Retained locks in case of Db2 member failure Member-Id
02
Resource Name Mode Status
102.2.A4 X ACTIVE
102.7.BA X RETAIN

DB2A 102.2.A5 U ACTIVE

IRLM
Member-Id 102.2.28 X ACTIVE
01 102.7 IX RETAIN

3
Db2 LOCK1 structure …
• LOCK1 structure size is defined in the CFRM policy
By default, Db2 tries to do a 50-50 split between
Lock Hash Table and Record List Table
Lock Hash Table (LTEs) = 256MB
STRUCTURE NAME(DSNDB2_LOCK1) Record List Table (RLEs) = 256MB
SIZE(1024M)
INITSIZE(512M) LTE size depends on number of data sharing members
ALLOWAUTOALT(NO) 2 bytes for 1-7 members
PREFLIST(COUPLE01,COUPLE02) 4 bytes for 8-23 members
8 bytes for 24-32 members

The Lock Hash Table has to be a power of 2

STRUCTURE NAME(DSNDB2_LOCK1)
SIZE(1024M) Lock Hash Table (LTEs) = 256MB
INITSIZE(768M) Record List Table (RLEs) = 512MB
ALLOWAUTOALT(NO)
PREFLIST(COUPLE01,COUPLE02) The record table is susceptible to storage shortages if the
structure is too small or if the allocation of the lock table
leaves too little storage for the record table

4
Db2 LOCK1 structure …
• Shortage of RLEs

DXR170I irlmx THE LOCK STRUCTURE wwwwwwww IS zz% IN USE  Alert at 50%, 60%, 70%
DXR142E irlmx THE LOCK STRUCTURE wwwwwwww IS zzz% IN USE  Alert at 80%, 90%, 100%

• At 80% full, data sharing continues with no restrictions, but storage is approaching a critical threshold
• At 90% full, only 'must-complete' type of requests that require lock structure storage are processed
• At 100% full, any request that requires lock structure storage is denied with an 'out of lock structure storage’ error

5
Global contention
ROT

DATA SHARING LOCKING

---------------------------
QUANTITY
--------
/SECOND
-------
/THREAD
-------
/COMMIT
-------
Global Contention should be less than 3-5% of XES IRLM Requests
GLOBAL CONTENTION RATE (%) 0.64
FALSE CONTENTION RATE (%) 0.11
... Global Contention = SUSPENDS – IRLM + XES + FALSE (A)
SYNCH.XES - LOCK REQUESTS 227.5M 10.6K 1368.75 458.86 XES IRLM Requests = (SYNCH. XES – LOCK + CHANGE + UNLOCK)
SYNCH.XES - CHANGE REQUESTS 1340.7K 62.24 8.07 2.70 + ASYNCH.XES – CONVERTED LOCKS + (SUSPENDS – IRLM + XES + FALSE) (B)
SYNCH.XES - UNLOCK REQUESTS 225.8M 10.5K 1358.14 455.30
ASYNCH.XES -CONVERTED LOCKS 3485.41 0.48 3.87 0.00
Global Contention Rate (%) = (A)/(B)*100
...
SUSPENDS - IRLM GLOBAL CONT 34192.00 1.59 0.21 0.07
SUSPENDS - XES GLOBAL CONT.
SUSPENDS - FALSE CONT. MBR
6076.00
21575.00
0.28
1.00
0.04
0.13
0.01
0.04 ROT

False Contention should be less than 1-3% of XES IRLM Requests

False Cont = false contention on lock table hash anchor point False Contention = SUSPENDS – FALSE (C)
Could be minimised by increasing the number of LTEs in the False Contention Rate (%) = (C)/(B)*100

Lock Hash Table

6
Resizing the Db2 LOCK1 structure
• Increase size of LOCK1 structure if high False Contention rate and/or shortage of RLEs
• If shortage of RLEs but False Contention rate is OK, consider using the IRLM option LTE= to control the
size of the Lock Hash Table
• Specify a value for the LTE= parameter in the IRLMPROC
or issue the MODIFY irlmproc SET,LTE= command
• Requires a REBUILD of LOCK1 structure
For LTE= 2-byte entries 4-byte entries
8 16 MB 32 MB
INITSIZE = 1280MB 16 32 MB 64 MB
LTE = 256 (based on 2-byte entries) 32 64 MB 128 MB
Lock Hash Table (LTEs) = 512MB 64 128 MB 256 MB
INITSIZE = 1024MB Record List Table (RLEs) = 768MB
Lock Hash Table (LTEs) = 512MB 128 256 MB 512 MB
Record List Table (RLEs) = 512MB 256 512 MB 1024 MB
INITSIZE = 1024MB
LTE = 128 (based on 2-byte entries) 512 1024 MB 2048 MB
Lock Hash Table (LTEs) = 256MB 1024 2048 MB 4096 MB
Record List Table (RLEs) = 768MB 2048 4096 MB

7
Global contention …
• Use a light-weight locking protocol (isolation level) and exploit lock avoidance
• Benefits of lock avoidance
• Increased concurrency by reducing lock contention
• Decreased lock and unlock activity and associated CPU resource consumption
• In data sharing, decreased number of CF requests and associated CPU overhead
• Minimise impact of retained locks
• Use ISOLATION(CS) CURRENTDATA(NO) or use ISOLATION(UR)
• Commit frequently
• Reduce lock contention
• Improve effectiveness of global lock avoidance
• Avoid serialisation points e.g. single row table used as a ‘counter’
• Use IDENTITY column or pull value from SEQUENCE object with CACHE
• Exploit table and index partitioning

8
Global contention …
ROT

P-lock Negotiation should be less than 3-5% of XES IRLM requests

DATA SHARING LOCKING QUANTITY /SECOND /THREAD /COMMIT
--------------------------- -------- ------- ------- -------
...
SYNCH.XES - LOCK REQUESTS 227.5M 10.6K 1368.75 458.86
P-lock Negotiation = PSET/PART P-LCK NEGOTIATION + PSET/PART P-LCK NEGOTIATION
SYNCH.XES - CHANGE REQUESTS 1340.7K 62.24 8.07 2.70 + OTHER P-LCK NEGOTIATION (A)
SYNCH.XES - UNLOCK REQUESTS 225.8M 10.5K 1358.14 455.30 XES IRLM Requests = (SYNCH. XES – LOCK + CHANGE + UNLOCK) + ASYNC.XES –
ASYNCH.XES -CONVERTED LOCKS 1315.00 0.06 0.01 0.00 CONVERTED LOCKS + (SUSPENDS – IRLM + XES + FALSE) (B)
... P-lock Negotiation Rate (%) = (A)/(B)*100
PSET/PART P-LCK NEGOTIATION 18037.00 0.84 0.11 0.04
PAGE P-LOCK NEGOTIATION 2863.00 0.13 0.02 0.01
OTHER P-LOCK NEGOTIATION 9067.00 0.42 0.05 0.02
P-LOCK CHANGE DURING NEG. 15991.00 0.74 0.10 0.03 • P-lock contention and negotiation can cause IRLM
latch contention, page latch contention,
asynchronous GBP write, active log write, GBP read
• Page P-lock contention by one thread causes Page Latch
contention for all other threads in the same member
trying to get to the same page

9
Global contention …
• Breakdown by page P-lock type in GBP statistics
For insert-intensive workloads with high page P-lock contention on space map
GROUP TOTAL CONTINUED QUANTITY /SECOND /THREAD /COMMIT pages, consider MEMBER CLUSTER (optionally combined with APPEND + LOCKSIZE
----------------------- -------- ------- ------- ------- ROW)
... Do not use APPEND or LOCKSIZE ROW without MEMBER CLUSTER
PAGE P-LOCK LOCK REQ 877.4K 122.88 14.91 3.64
for an INSERT-at-the-end intensive workload  may result in excessive
SPACE MAP PAGES 83552.00 11.70 1.42 0.35
DATA PAGES 10663.00 1.49 0.18 0.04 page p-lock contention on data pages and space map pages
INDEX LEAF PAGES 783.2K 109.69 13.31 3.25
For heavily concurrent insert activity (many concurrent threads) with high page
PAGE P-LOCK UNLOCK REQ 926.8K 129.80 15.75 3.84 P-lock contention on data pages caused by space search and false leads, consider
PAGE P-LOCK LOCK SUSP 8967.00 1.26 0.15 0.04
INSERT ALGORITHM 2 (aka Fast Un-clustered INSERT)
SPACE MAP PAGES 593.00 0.08 0.01 0.00
DATA PAGES 143.00 0.02 0.00 0.00 If data page P-lock contention on small tables with LOCKSIZE ROW, consider
INDEX LEAF PAGES 8231.00 1.15 0.14 0.03 MAXROWS=1 and LOCKSIZE PAGE to ‘simulate’ row locking and reduce spacemap
free space update
PAGE P-LOCK LOCK NEG 10285.00 1.44 0.17 0.04
SPACE MAP PAGES 8.00 0.00 0.00 0.00
DATA PAGES 10.00 0.00 0.00 0.00 If index tree P-lock (high index splits), consider
INDEX LEAF PAGES 10267.00 1.44 0.17 0.04 • Freespace tuning (PCTFREE, FREEPAGE)
• Exploit data and index partitioning to ‘dilute’ hot spot
• Increase index page size – warning: could also aggravate contention!

10
Global contention …
CLASS 3 SUSPENSIONS AVERAGE TIME AV.EVENT TIME/EVENT
• Db2 accounting for more granular information --------------------
LOCK/LATCH(DB2+IRLM)
------------
0.000097
--------
0.58
----------
0.000167
IRLM LOCK+LATCH 0.000004 0.21 0.000021
DB2 LATCH 0.000092 0.37 0.000251
..
PAGE LATCH 0.000595 1.89 0.000314
NOTIFY MSGS 0.000000 0.00 N/C
GLOBAL CONTENTION 0.004844 9.26 0.000523

GLOBAL CONTENTION L-LOCKS AVERAGE TIME AV.EVENT GLOBAL CONTENTION P-LOCKS AVERAGE TIME AV.EVENT
------------------------------------- ------------ -------- ------------------------------------- ------------ --------
L-LOCKS 0.000011 0.05 P-LOCKS 0.004833 9.21
PARENT (DB,TS,TAB,PART) 0.000000 0.00 PAGESET/PARTITION 0.000000 0.00
CHILD (PAGE,ROW) 0.000000 0.00 PAGE 0.004790 9.16
OTHER 0.000011 0.05 OTHER 0.000043 0.05

LOCKING AVERAGE TOTAL DATA SHARING AVERAGE TOTAL GROUP TOT4K AVERAGE TOTAL
--------------------- -------- -------- ------------------- -------- -------- --------------------- -------- --------
... GLOBAL CONT RATE(%) 3.88 N/A ...
LOCK REQUEST 139.05 2642 FALSE CONT RATE(%) 0.14 N/A PG P-LOCK LOCK REQ 65.26 1240
UNLOCK REQUEST 34.63 658 ... SPACE MAP PAGES 6.95 132
QUERY REQUEST 56.26 1069 LOCK REQ - XES 128.00 2432 DATA PAGES 16.11 306
CHANGE REQUEST 13.32 253 UNLOCK REQ - XES 78.21 1486 INDEX LEAF PAGES 42.21 802
OTHER REQUEST 0.00 0 CHANGE REQ - XES 5.63 107 PG P-LOCK UNLOCK REQ 58.26 1107
TOTAL SUSPENSIONS 0.32 6 SUSPENDS - IRLM 8.26 157 PG P-LOCK LOCK SUSP 8.84 168
LOCK SUSPENSIONS 0.00 0 SUSPENDS - XES 0.00 0 SPACE MAP PAGES 0.95 18
IRLM LATCH SUSPENS. 0.32 6 CONVERSIONS- XES 0.68 13 DATA PAGES 1.84 35
OTHER SUSPENS. 0.00 0 FALSE CONTENTIONS 0.32 6 INDEX LEAF PAGES 6.05 115

11
Global contention …
• Lock suspension report to identify ‘hot spots’
• For detailed analysis, start the following Db2 Performance traces for short periods of time during peak processing
• TRACE(P) CLASS(30) IFCID (44,45,105,107,213,214,215,216,226,227)
• Sample Db2 OMPM/PE report to generate a CSV file that can be easily loaded into a spreadsheet

LOCKING
REPORT
LEVEL(SUSPENSION)
DDNAME(LORPTDD1)
SPREADSHEETDD(SPSHDD)
ORDER(DATABASE-PAGESET)

12
Db2 group buffer pool structures
• What are they used for?
• Register buffers for cross-invalidation (XI)
• ‘List’ option provided for prefetch Db2 GBP Structure
DB2B UPDATERS
• Write changed buffers, send XI signals WRITE Group Buffer Pool Directory
Local Cache Vector
• H/W instruction to test vector bits for buffer XI UPDATED
• Fast refresh of XI'ed buffers BPn
PAGES
P2
• Store-in cache by default
• ‘No-data’ option provided P2
• Force-at-commit database write
DB2A
protocol used for writes to CF
Local Cache Vector

BPn READERS
P1 REGISTER
INTEREST
Group Buffer Pool Cache

13
Cross invalidations
• Two reasons for cross invalidations
• Perfectly normal condition in an active-active data sharing environment
• Directory entry reclaims – condition you want to tune away from
• CPU overhead and I/O overhead if there is not enough directory entries
• Extra CF access and Sync I/O Read
• -DISPLAY GROUPBUFFERPOOL(*) TYPE(GCONN)

DSNB787I - RECLAIMS
FOR DIRECTORY ENTRIES = 0
FOR DATA ENTRIES = 0 ROT 1/3
CASTOUTS = 0

DSNB788I - CROSS INVALIDATIONS

DUE TO DIRECTORY RECLAIMS = 0 Reclaims for Directory Entries should be minimised
DUE TO WRITES = 0
EXPLICIT = 0 Cross Invalidations due to Directory Reclaims should be zero

14
GBP reads

ROT 2/3
GROUP BP14 QUANTITY /SECOND /THREAD /COMMIT
----------------------------- -------- ------- ------- -------
...
SYN.READ(XI)-DATA RETURNED 1932.00 0.09 0.01 0.00 Sync.Read(XI) miss ratio should be < 10%
SYN.READ(XI)-NO DATA RETURN 39281.6K 1823.66 236.31 79.22
SYN.READ(NF)-DATA RETURNED 22837.00 1.06 0.14 0.05
SYN.READ(NF)-NO DATA RETURN 6955.8K 322.93 41.85 14.03 TOTAL SYN.READ(XI) (A) = SYN.READ(XI)-DATA RETURNED
+ SYN.READ(XI)-NO DATA RETURN
Sync.Read(XI) miss ratio = SYN.READ(XI)-NO DATA RETURN / (A)

• Local BP search  GBP search  DASD I/O

• SYN.READ(NF) = Local Buffer Pool miss
• SYN.READ(XI) = Local Buffer Pool hit but cross-invalidated buffer
• Most data should be found in GBP  if not, GBP may be too small or pages have been removed because of
directory entry reclaims

15
GBP writes
• Changed Pages Sync Written to GBP / force write GROUP BP14 CONTINUED QUANTITY /SECOND /THREAD /COMMIT
• Commit -----------------------
WRITE AND REGISTER
--------
54896.00
-------
2.55
-------
0.33
-------
0.11
• P-lock negotiation WRITE AND REGISTER MULT
CHANGED PGS SYNC.WRTN
255.5K
408.3K
11.86
18.96
1.54
2.46
0.52
0.82

• Changed Pages Async Written to GBP CHANGED PGS ASYNC.WRTN

...
1713.4K 79.55 10.31 3.46

• Deferred write PAGES IN WRITE-AROUND 0.00 0.00 0.00 0.00

• System checkpoint
• Pages in Write-Around
• Applies only to Pages Async Written to GBP
• Db2 conditionally enables & disables the GBP write-around protocol
• Turned on at GBP level threshold 50%, GBP Class threshold 20%
• Turned off at GBP level threshold 40%, GBP Class threshold 10%
• Pages are written directly to DASD instead of to the GBP
• Cross invalidation signals sent to local BPs after DASD write I/O is complete

16
GBP castout
• GBP castout thresholds are similar to local BP deferred write thresholds
• Encourage Class_castout (CLASST) threshold by lowering its value
• More efficient than GBP_castout threshold (notify to pageset/partition castout owner)
• CLASST threshold check by GBP write
• GBPOOLT threshold check by GBP castout timer (10sec default)
• Default thresholds lowered in V8
• V11: Class castout threshold below 1%

V7 V8/V9/V10/V11
GROUP BP14 QUANTITY /SECOND /THREAD /COMMIT
VDWQT (dataset level) 10% 5%
----------------------------- -------- ------- ------- -------
PAGES CASTOUT 2224.8K 103.28 13.38 4.49 DWQT (buffer pool level) 50% 30%
UNLOCK CASTOUT 58868.00 2.73 0.35 0.12
... CLASST (Class_castout) 10% 5%
CASTOUT CLASS THRESHOLD 26835.00 1.25 0.16 0.05
GROUP BP CASTOUT THRESHOLD 594.00 0.03 0.00 0.00 GBPOOLT (GBP_castout) 50% 30%
GBP CHECKPOINTS TRIGGERED 45.00 0.00 0.00 0.00
GBPCHKPT (GBP checkpoint) 8 4

17
GBP castout …
• As transaction and data volumes grow, the GBP can become stressed
• Pages written to GBP faster than castout engines can process
• Group buffer pool congested with changed pages
• Can cause group buffer pool full condition in extreme cases
21.02.15 S0012052 *DSNB325A -DP1A DSNB1CNE THERE IS A CRITICAL SHORTAGE OF SPACE IN GROUP BUFFER POOL GBP11
...
21.07.37 S0012052 DSNB327I -DP1A DSNB1CNE GROUP BUFFER POOL GBP11 HAS ADEQUATE FREE SPACE

• Problems are often precipitated by update-intensive batch jobs or utilities run against GBP dependent
objects
• Intense, sustained GBP page write activity can lead to a shortage of GBP memory
• Automatic GBP ALTER via XES Autoalter can respond and increase GBP size up to SIZE
• Provided there is sufficient headroom …

18
GBP castout …
• GBP shortages may impact application performance and ultimately become an availability exposure
• When the GBP fills up, Db2 starts pacing for the commit - slower response time
DSNB750I -DP11 DISPLAY FOR GROUP BUFFER POOL GBP11 FOLLOWS GROUP BP14 QUANTITY /SECOND /THREAD /COMMIT
... ----------------------------- -------- ------- ------- -------
DSNB786I -DP11 WRITES CASTOUT CLASS THRESHOLD 26835.00 1.25 0.16 0.05
CHANGED PAGES = 688259863 GROUP BP CASTOUT THRESHOLD 594.00 0.03 0.00 0.00
CLEAN PAGES = 0 GBP CHECKPOINTS TRIGGERED 45.00 0.00 0.00 0.00
FAILED DUE TO LACK OF STORAGE = 3630 WRITE FAILED-NO STORAGE 0.00 0.00 0.00 0.00
CHANGED PAGES SNAPSHOT VALUE = 44474

• After repetitive write failures page will be put on the LPL

• If the failures are against an index, the entire index might be put on the LPL
• Recovery actions are then necessary
Reduce castout thresholds, and/or
ROT 3/3 WRITE FAILED-NO STORAGE < 1% of TOTAL CHANGED PGS WRTN Reduce GBP checkpoint timer and/or
Increase GBP size

19
Group buffer pool tuning
• -DIS GBPOOL(*) GDETAIL(*) TYPE(GCONN)
DSNB750I -PR4B DISPLAY FOR GROUP BUFFER POOL GBP2 FOLLOWS
... Note:
DSNB757I -PR4B MVS CFRM POLICY STATUS FOR DSNPR0B_GBP2 = NORMAL
MAX SIZE INDICATED IN POLICY = 614400 KB Make sure to collect Statistics
... Class 5 (IFCID 230)
DSNB758I -PR4B ALLOCATED SIZE = 614400KB  Additional GBP stats
...
DSNB759I -PR4B NUMBER OF DIRECTORY ENTRIES = 384147 including Reclaims and XIs
NUMBER OF DATA PAGES = 116010
... OMPE stores the GBP stats in
DSNB786I -PR4B WRITES
CHANGED PAGES = 2882842576 two different tables:
CLEAN PAGES = 0 DB2PM_STAT_GBUFFER
• FAILED DUE TO LACK OF STORAGE = 71 DB2PMSYSPAR_230
CHANGED PAGES SNAPSHOT VALUE = 10642
DSNB787I -PR4B RECLAIMS
FOR DIRECTORY ENTRIES = 2495178
FOR DATA ENTRIES = 3290663329
CASTOUTS = 4018446743
DSNB788I -PR4B CROSS INVALIDATIONS
DUE TO DIRECTORY RECLAIMS = 586960
DUE TO WRITES = 877165837
EXPLICIT = 2
20
Group buffer pool tuning …
• GBP size is defined in CFRM policy

STRUCTURE NAME(DSNDB2_GBP1)
SIZE(1024M)
INITSIZE(512M)

• Additionally, you can specify a RATIO using the ALTER GROUPBUFFERPOOL command to indicate how
many Directory Entries (ENTRIES) per Data Pages (ELEMENTS)

APAR PH13045 introduces 2 changes for Db2 12 users: NEW

Default value of RATIO: 5  10
Limit of RATIO on the ALTER GROUPBUFFERPOOL command: 255  1024

21
Group buffer pool tuning …
• Example based on DISPLAY GBPOOL output
SUM VPSIZEs +
GBPOOL SIZE (MB) ALLOC_SIZE (MB) DIR_ENTRIES DATA_PAGES RATIO (1) DATA_PAGES
GBP0 150 90 74285 14857 5 34857
GBP1 400 300 318108 45444 7 225444
GBP2 600 550 225969 116010 1.9 356010
GBP8K0 990 990 403011 106257 3.8 456257
GBP8K1 600 500 554987 36998 15 436998
GBP16K0 120 60 15792 3156 5 5156
GBP32K 220 120 48499 3030 16 73030

GBPOOL % COVERAGE FAIL_LACK_OF_STG DIR_RECLAIMS XI_DIR_RECLAIMS

GBP0 213% 0 0 0
GBP1 141% 0 0 0
GBP2 63% 71 2495178 586960
GBP8K0 88% 4231 35733124 12267527
GBP8K1 127% 0 0 0
GBP16K0 306% 0 0 0
GBP32K 66% 0 0 0

22
Group buffer pool tuning …

• Targeted tuning #1: GBP with large number of directory reclaims and XI due to directory reclaims (but no
or minimal write failures)  GBP2
• Tuning:
• Keep number of data pages the same to avoid aggravating write failures
• Increase SIZE and RATIO to cover the max number of directory entries that could ever be required (1 for each
local buffer + 1 for each GBP data page)
• Note: RATIO can be a decimal value with 1 digit after the decimal point (e.g. 5.2)
• Changes required:
NEW DIR ENTRIES = SUM VPSIZE across all Db2 members + GBP DATA PAGES
NEW RATIO = NEW DIR ENTRIES / GBP DATA PAGES
NEW INITSIZE = (GBP DATA PAGES * PAGE SIZE (KB) + NEW DIR ENTRIES * 410 bytes / 1024) / 1024
NEW SIZE = 1.3 to 2x NEW INITSIZE

The size of a Directory Entry can vary by CF level

For a rough estimate, use 410-430 bytes per entry on CF Level 17
23
Group buffer pool tuning …
• Targeted tuning #2: GBP with large number of write failures (with or without large number of directory
reclaims and XI due to directory reclaims)  GBP8K0
• Tuning:
• Adjust CLASST and GBPOOLT to trigger more frequent castout
• Increase the number of data pages to reduce critical space shortages and write failures
• Adjust size and ratio to still cover the max number of directory entries that could ever be required (1 for each
local buffer + 1 for each GBP data page)
• Note: RATIO can be a decimal value with 1 digit after the decimal point (e.g. 5.2)
• Changes required:
NEW GBP DATA PAGES = 1.3 to 2x GBP DATA PAGES
NEW DIR ENTRIES = SUM VPSIZE across all Db2 members + NEW GBP DATA PAGES
NEW RATIO = NEW DIR ENTRIES / NEW GBP DATA PAGES
NEW INITSIZE = (NEW GBP DATA PAGES * PAGE SIZE (KB) + NEW DIR ENTRIES * 410 bytes / 1024) / 1024
NEW SIZE = 1.3 to 2x NEW INITSIZE

24
Group buffer pool tuning …
• Simplified tuning when using XES AUTO ALTER
• Very useful autonomic functionality to simplify GBP tuning
• Tries to avoid Structure Full conditions
• Tries to avoid Directory Reclaim conditions
• Recommendations:
• Set CFRM policy properties
• ALLOWAUTOALT(YES)
• FULLTHRESHOLD = 80-90%
• SIZE = 1.3-2x INITSIZE
• MINSIZE = INITSIZE
• Periodically review GBP actual allocations
• If SIZE is reached, it limits the effectiveness of XES AUTO ALTER  plan to update the CFRM policy and schedule
a REBUILD to increase the structure

25
Group buffer pool tuning …
• Simplified tuning when using XES AUTO ALTER …

• Changes when allocated size reaches SIZE - but no or minimal write failures
NEW DIR ENTRIES = SUM VPSIZE across all Db2 members + GBP DATA PAGES
NEW RATIO = NEW DIR ENTRIES / GBP DATA PAGES
NEW INITSIZE = (GBP DATA PAGES * PAGE SIZE (KB) + NEW DIR ENTRIES * 410 bytes / 1024) / 1024 A more sophisticated approach
NEW SIZE = 1.3 to 2x NEW INITSIZE would be to look at the peak
rate of Changed Pages written
to the GBP and calculate the
• Changes when allocated size reaches SIZE with large number of write failures average residency time 
NEW GBP DATA PAGES = 1.3 to 2x GBP DATA PAGES tuning target 30-60 seconds
NEW DIR ENTRIES = SUM VPSIZE across all Db2 members + NEW GBP DATA PAGES
NEW RATIO = NEW DIR ENTRIES / NEW GBP DATA PAGES
NEW INITSIZE = (NEW GBP DATA PAGES * PAGE SIZE (KB) + NEW DIR ENTRIES * 410 bytes / 1024) / 1024
NEW SIZE = 1.3 to 2x NEW INITSIZE

26
Group buffer pool tuning …
• Increase of local buffer pool size on a ‘healthy GBP’

• Tuning:
• Keep number of data pages the same to avoid aggravating write failures
• Increase size and ratio to cover the max number of directory entries that could ever be required (1 for each
local buffer + 1 for each GBP data page)

• Changes required:
NEW DIR ENTRIES = SUM VPSIZE across all Db2 members + GBP DATA PAGES
NEW RATIO = NEW DIR ENTRIES / GBP DATA PAGES
NEW INITSIZE = (GBP DATA PAGES * PAGE SIZE (KB) + NEW DIR ENTRIES * 410 bytes / 1024) / 1024
NEW SIZE = 1.3 to 2x NEW INITSIZE

27
Questions?

28