Upload
others
View
12
Download
0
Embed Size (px)
Citation preview
What's new in EPV for Db2 and
EPV Graph for Db2 V15
Massimo Orlando
EPV Technologies
XIX EPV User Group
Agenda
EPV for Db2 enhancements
EPV Graph for Db2 enhancements
New version plan
2
XIX EPV User Group
Introduction
In this presentation we will focus on some of the most relevant
enhancements introduced in EPV for Db2 and EPV Graph for
Db2 V15
Many other enhancements have been introduced but they will
not be discussed here
This version provides full support of z15 and Db2 12
It also fully supports the new EPV UI
4
XIX EPV User Group
EPV for Db2 enhancements
Continuous delivery information
Fast Traverse Block
Accelerator exploiters
Plan Unlock/Commit Ratio
6
XIX EPV User Group
Continuous Delivery
In Db2 V12, IBM adopted the continuous delivery model
The idea is to deliver enhancements in the service stream without
waiting for a massive new release
Continuous delivery should solve two main customer issues:
▪ Customers saw deployment of new releases as disruptive,
making them reluctant to move to a new release, which inhibited
their ability to take advantage of new features
▪ Many customers wanted new functions delivered much faster (not
every 3 years as it happened with Db2 V11 and before)
7
XIX EPV User Group
With continuous delivery it’s very important being able to check
the levels of Db2 code, catalog and functions used
This information is not provided in any SMF records
It can be easily collected by running a JCL step which executes
the DISPLAY GROUP command, on each Db2 subsystems
This JCL is provided in EPV for Db2 and it should be already
used in previous versions to collect Db2 parameters and
statistics
8
Continuous Delivery
XIX EPV User Group
Fast Traverse Block
Many Db2 applications rely heavily on one or more indexes on each
table in order to locate specific data rows efficiently
Db2 indexes use a classic “b-tree” structure with a root page,
usually one or more levels of non-leaf pages and leaf pages pointing
to the data pages
The more levels of index, the higher the cost in terms of the number
of get pages, consequent CPU cycles required and elapsed time
If the index levels don’t reside in a buffer pool, there will be an
additional penalty due to the required I/O operations
12
XIX EPV User Group
Fast Traverse Block
To perform more efficient random index lookups Db2 12
introduces the index Fast Traverse Block (FTB) function
With FTB, Db2 will cache root and non-leaf page information in
the FTB storage pool but only for UNIQUE indexes with a key
size of 64 bytes or less
Db2 will automatically and dynamically decide which indexes to
cache based on current workload access patterns
13
XIX EPV User Group
A new Db2 storage pool has been created in V12 to host the FTBs; its size is determined by the INDEX_MEMORY_CONTROL parameter
Possible values are:
▪ AUTO; Db2 will size the FTB storage pool to the maximum between 20% of the total allocated buffer pool size and 500MB; it’s the default
▪ DISABLE; Db2 will disable the FTB functionality
▪ A value in the 500-200.000 MB range; Db2 will size the FTB storage pool based on that value
When in data sharing it’s recommended that all members use the same setting
14
Fast Traverse Block
XIX EPV User Group
Fast Traverse Block 15
FTB Usage
▪ Indexes, is the number of indexes with FTB blocks
▪ Size, is the total FTB size in bytes
XIX EPV User Group
Fast Traverse Block 16
FTB Usage
▪ Index Levels, is the number of index levels with FTB blocks
▪ Size, is the total FTB size in bytes
XIX EPV User Group
Accelerator exploiters
EPV for Db2 already provided information about accelerator
utilization …
17
XIX EPV User Group
Accelerator exploiters
… and accelerator activity in previous versions
More details are available by clicking the server id link
18
XIX EPV User Group
In EPV for Db2 V15 we introduced views which allows to
analyze the TOP accelerator exploiters:
▪ by CPU time
▪ by elapsed time
19
Accelerator exploiters
XIX EPV User Group
Plan Unlock/Commit Ratio
Lock avoidance is a combination of techniques used by Db2 to
retrieve data without requiring holding a shared lock on behalf of the
application process, if some conditions are met
In this case, Db2 will take a page latch (and page p-lock in data
sharing) to ensure physical consistency of the page instead of a lock
It only applies to low level locks (page and row)
The application has no direct control over it
But applications not committing frequently enough may disrupt it
20
XIX EPV User Group
Plan Unlock/Commit Ratio
An Unlock/Commit Ratio (UCR) greater than 5 will indicates
that lock avoidance is not exploited well
EPV for Db2 already provided this info at the subsystem level
21
LOCK ACTIVITY 7 8 9 10 11 12 13 14 15 16 17 18 19LOCK REQ/SEC 17.955 16.285 15.191 37.449 40.204 31.742 25.629 28.229 29.254 18.243 13.554 32.555 25.774
LOCK CHGD/SEC 851 2.132 864 590 525 518 474 569 497 373 355 718 727
LOCK QRY/SEC 0 0 0 0 0 0 0 0 0 0 0 0 0
UNLOCK/SEC 5.880 4.434 4.637 21.646 11.656 8.042 6.171 6.660 8.060 4.303 3.346 5.918 5.868
COMMIT/SEC 698 934 453 491 408 440 346 377 406 358 428 639 970
UNLOCK/COMMIT 8 5 10 44 29 18 18 18 20 12 8 9 6
LOCK SUSP/SEC 9 20 13 10 7 8 7 9 8 7 8 8 11
IRLM LATCH SUSP/SEC 305 318 108 219 201 201 121 192 224 114 197 3.437 1.283
OTHER SUSP/SEC 4 2 2 4 2 3 3 2 2 2 3 5 3
IRLM LATCH CONTENTION 1 1 1 0 0 0 0 1 1 0 1 9 4
XIX EPV User Group
Plan unlock/commit ratio
In EPV for Db2 V15 we calculate and publish the UCR for the
TOP transactions (CICS, IMS), authids (DDF, SAF) and plans
22
XIX EPV User Group
Latch analysis
In EPV Graph for Db2 V15 we added graphs showing the daily
profile and the daily trends of the latch rate at data sharing
group subsystem and latch class level
All the latch classes (more than 30) are reported
We also provide monthly trends for long term analysis
25
XIX EPV User Group
EDM pools efficiency
In EPV Graph for Db2 V15 we added graphs showing the daily
profile and the daily trends of the hit ratio of:
▪ DBD pool; it should be close to 100%
▪ SKELETON pool; Cursor Tables for plans and Package Tables
for packages; they should also be close to 100%
▪ Global Dynamic Statement Cache; it should be more than 95%
Also in this case, we provide monthly trends for long term
analysis
29
XIX EPV User Group
EDM pools efficiency
Daily trends of
the skeleton
cursor tables
hit ratio in the
different time
shifts
30
XIX EPV User Group
EDM pools efficiency
Daily trends of the skeleton package tables hit ratio
Only prime shift has been selected in the graph
We also zoomed the graph (see Y axis)
31
XIX EPV User Group
New version plan
EPV for Db2 and EPV Graph for Db2 will continue to evolve
V16 is planned by mid 2022
It will include all V16 new features and enhancements (see
EPV for z/OS V16 preview)
It will support z16 (?) machines (expected in Spring 2022)
It will continue to support new Db2 function levels and versions
33