29
DB2 Performance Tuning XVI EPV USER GROUP Rome 11 October 2018 ROBERTO GIOI Responsabile Capacity ed Ottimizzazione

DB2 Performance Tuning - epvtech.com

  • Upload
    others

  • View
    11

  • Download
    0

Embed Size (px)

Citation preview

DB2 Performance Tuning XVI EPV USER GROUP

Rome 11 October 2018

ROBERTO GIOI

Responsabile Capacity ed Ottimizzazione

2

MPS and EPV

• MPS has been using EPV software since 2006. Started as support tool to the

sysprogs, from 2009 in Capacity Management Dept.

• Products installed: EPV zParser, EPV for z/OS, EPV Graph for z/OS and

EPV for Db2

• Custom application developed by EPV - EPVMPS - for business accounting

reports purposes

• All the performance & tuning tasks examined in this document rely on

multiple metrics, most of them provided by EPV products

• The tables presented are snapshots of the standard EPV HTML tables

• All the graphs are created starting from the data collected in EPV DBs. MPS

has built an extensive set of self-updating daily excel reports from EPV DBs

3

MPS and EPV

4

MPS and EPV

5

Overview of performance & tuning tasks

6

First phase:

RSU application

ZPARM refresh

Rebind

High Performance DBAT

7

Findings on main Production Subsys…

• Increase DB2 address space priority in DBSTC + IRLM in SYSTC

• Reduce service class periods for consistency

• Perform Rebind in order to remove invalid/bypassed S-proc

• Increase Plan and Package Auth Cache (AUTHCACHE, CACHEPAC)

• Increase memory in all systems (several go low)

• Increase CF structures size in order to reduce/eliminate reclaims

• Many Buffer Pools to be Page-fixed

• Huge Buffer Pools (>6M pages) to be split up

• Set TRACKMODE=NO

• Define more/larger 4K and 32K workfiles

8

…and others

• Increase DB2 address space priority in DBSTC + IRLM in SYSTC

• Reduce service class periods for consistency

• Set consistent BP definition within data sharing groups

• Increase log buffers (OUTBUFF)

• Increase EDM_SKELETON _POOL

• Increase MAXDS

• Increase NUMLKTS e NUMLKUS

• Disable REAL_STORAGE_MANAGEMENT in all DB2s (CRIT_SIT)

• Put Workfiles and DGTT in separate BP from other objects

• Set VDWT by number of buffers rather than by % of BP size

9

Deployment schedule for main DB2 subsys

LPAR: OS0A

· 14 January RSU application, ZPARM, Rebind

· 24 January 18:00 High Performance DBAT

LPAR: OS0U

· 28 January RSU application, ZPARM, Rebind, 2 new BPs

· 2 February 18:00 High Performance DBAT

LPAR: CICS e PROD

· 4 February RSU application, ZPARM, Rebind, 2 new BPs

· 18 February 02:00 High Performance DBAT

10

DB2 address spaces findings

LPAR: OS0A

RSU application, ZPARM, Rebind - Prime Time shift:

DB2 Master CPU -20%

High Performance DBAT enablement - Prime Time shift:

average DDF call single-execution CPU -20%

LPAR: OS0U

RSU application, ZPARM, Rebind, 2 new BPs - in Prime Time shift:

DB2 Master CPU -22%

High Performance DBAT enablement - Prime Time shift:

average DDF call single-execution CPU -6,5%

11

DB2 subsys results: OS0A and OS0U LPARs

12

DB2 subsys results: CICS LPAR

· RSU application, ZPARM, Rebind - Prime Time shift:

DB2 Master CPU -26% ~ -53 mips/hour usage

13

DB2 subsys results: LPAR CICS

High Performance DBAT enablement - Prime Time Shift:

average DDF call single-execution CPU -10%

on 28 million transactions/hour ~ -250 mips/hour usage

14

CPI per LPAR: CICS

15

Second phase:

SMT enablement

16

SMT CICS and OS0U LPARs

In the 8-18 hour shift (full day), zIIP Eligible average mips usage (aka

IIPCP: processor time used by zIIP eligible transactions running on

general purpose processors) decreased by about 36 mips and the

ratio between zIIP Eligible vs. zIIP Executed dropped from 1,3% to

0,4%

During zIIP peak hours, zIIP Eligible dropped by 140 mips and the

ratio between zIIP Eligible/zIIP Executed dropped from 3,9% to 0,4%

At hour 8, opening time for the Bank’s branches, zIIP Eligible

decreased by about 250 mips on CICS-OS0U LPARs

17

SMT CICS and OS0U LPARs

18

SMT CICS and OS0U LPARs

19

SMT LPAR DBM2

In the 8-18 hours shift (full day), zIIP Eligible average mips usage (aka

IIPCP: processor time used by zIIP eligible transactions running on

general purpose processors) decreased by over 100 mips

The ratio between zIIP Eligible/zIIP Executed dropped from 10% to 2%

20

SMT LPAR DBM2

21

SMT LPAR DBM2

22

Conclusions

More than 300 mips consumption reduction in Prime Time Shift

Better and more stable response time and throughput

23

Conclusions

25

Effetti di REAL_STORAGE_MANAGEMENT

Migration from

DB2 v.10 to

DB2 v.11 CM

26

What’s

that spike

?

Geolocation SW

deployed

Effetti di REAL_STORAGE_MANAGEMENT

27

Migration from

DB2 v.10 to

DB2 v.11 CM

Geolocation

SW deployed

Effetti di REAL_STORAGE_MANAGEMENT

28

REALSTORAGE_MANAGEMENT in macro DSN6SPRM

The REALSTORAGE_MANAGEMENT subsystem parameter specifies whether DB2® should manage real storage

consumption.

Acceptable values: ON, OFF, AUTO

Default: AUTO

DSNZPxxx: DSN6SPRM REALSTORAGE_MANAGEMENT

ON

DB2 always discards unused real storage frames. Discarding the frames results in some CPU overhead, and this option

is intended for systems in which the availability of real storage is limited. This value would most likely be appropriate for

LPARs that have many DB2 subsystems, such as a development LPAR.

OFF

DB2 does not discard unused real storage frames until one of the following conditions is met:

The LPAR had reached an auxiliary critical state.

The total real and auxiliary storage has reached 80% of the value of the REALSTORAGE_MAX subsystem parameter.

AUTO

DB2 discards unused real storage frames when a significant amount of paging activity is detected. By discarding frames,

DB2 tries to bring the system to a point where paging is limited or nonexistent. However, it might not be possible to bring

the system to that point if other applications on the same LPAR cause the shortage of real storage frames. AUTO is the

default value.

Spatial function responsible Two PTFs: UI36156 UI37237

Effetti di REAL_STORAGE_MANAGEMENT

29

MSTR average CPU before 670 mips MSTR average CPU after 210 mips

-68%

Effetti di REAL_STORAGE_MANAGEMENT