Tuning group buffer pools
To tune the group buffer pools
On the Data Sharing Wizard (WZDSHAR), select Review Group (Session) to see an overview of key statistics about:
- Global contention
- Group buffer pool performance
- GBP-dependent page sets
- Coupling facility activity (castouts, failures)
>W1 =WZDSGMEM=========(DBGHC====*=======)10MAR2011==16:33:40====MVDB2====D====2
Db2 MVS Global Total Read Hit % GBP
Target System Cont. % Lockouts ..... 0...50..100 DEP Castouts Failures
DB1HC SYSC 3.76 4 14.0 ** 1 4 0
Db2HC SYSC 1.57 6 22.2 ** 1 23 0This view shows you some of the key indicators per member and lets you quickly identify whether you have any performance issues to analyze. The counts shown are for the time since Db2 was started.
From here, you can hyperlink on any Db2 to see all group buffer pool statistics for both the current interval and the session data since Db2 startup. Alternatively, you can return to WZDSHAR to follow some of the analysis paths.
- Press F3 to return to the Data Sharing Wizard (WZDSHAR)
Select Check GBPs.The GBP analysis section of the Data Sharing Wizard is displayed. This decision panel provides options to analyze GBP size and the ratio of directory to data entries.
W1 =WZDSBOPT=========(DBGHC====DB******)10MAR2011==16:35:19====MVDB2====D====1
** DATA SHARING WIZARD - GBPs Group GBP Values DBGK
Check Totals per GBP
(Select GBP for Detail) Interval Session
. GBP Size Too Small? Read Hit %...... 0.0 6.2
(Low Hit% / Cache Full) Miss-Cache Full. 0 0
Curr. Chngd Pgs. 0
. Too Few Directory Entries? Dir.Reclaims.... 0 0
(Reclaims Cause XI/DASD Reads) XIs/Dir.Recl.... 0 0
. Too Few Data Entries?
(Castouts More Frequent) Castouts........ 6 95
(Castout Too Slow) Write Fail-Stor. 0 0Three diagnostic paths are available. For this session, you are going to follow the path to analyze directory entries. Even if the pools are large enough, a lack of directory entries (used to register each page) can cause problems.
The key indicators on the right help you decide whether you need to do additional analysis by following one or more paths. On this panel, all of these values are for the whole group, summarized for all GBPs.
Possible symptoms of too few directory entries are:
- The occurrence of directory reclaims so that new pages can be registered
- Cross-invalidations of pages in the members’ local buffer pools, occurring because of these directory reclaims
If one of these indicators is greater than 0, the field is highlighted in red.
Hyperlink on Too Few Directory Entries?.A tabular list of all defined group buffer pools is displayed.
>W1 =WZDSGBR==========(DBGHC====*=======)10MAR2011==16:36:09====MVDB2====D====9
GBP Curr. Pend. Dir. Data Directory XI from XI Read Read Hit
Name Ratio Ratio Entries Entries Reclaims Dir.Recl. Miss Ratio
BP0 5 5 942 187 0 0 3 6.2
BP1 0 0 0 0 0 0.0
BP2 0 0 0 0 0 0.0
BP4 0 0 0 0 0 0.0
BP5 0 0 0 0 0 0.0
BP9 0 0 0 0 0 0.0
BP11 0 0 0 0 0 0.0
BP32K 0 0 0 0 0 0.0
BP32K9 0 0 0 0 0 0.0Now you can see each group buffer pool with the current definitions, the two key indicators that you saw before for all pools (Directory Reclaims and XI from Dir. Recl. columns), and some additional related statistics.
Select a single GBP for further analysis.
W1 =WZDSGBRD=========(DBGHC====DB******)10MAR2011==16:37:11====MVDB2====D====1
GBP Directory Entry Analysis - Detail BP0
Interval Session
Reclaims for Directory Entries? 0 0
(GBP Page/Dir Reused)
Directory Reclaims Causing XI? 0 0
(Local VP Pages Invalidated)
Also Increasing DASD Reads?
. Check Members for:
* XI Miss ==> DASD Reads
GBP Hit Ratio Low?
. Check Members for:
* Low GBP Hit Ratio
Definitions
Total Size (4K Blocks) 256
Directory Entries 942
Data Entries 187
Current Ratio 5
Pending Ratio 5This analysis panel provides details to help you determine whether a group buffer pool might have too few directory entries to support its part of the data sharing workload. This panel addresses the main issue—whether directory reclaims are causing cross-invalidations (XI) and probably increasing I/O.
In this panel, you now see the two key indicators again, but as interval (current activity) and session (total since Db2 startup) counts.
The answer to the additional question of whether this situation is increasing DASD reads can be seen only by looking at statistics per member instead of at the group level.
Hyperlink on Check Members for XI Miss.A view containing data sharing session activity counts per member for this group buffer pool is displayed.
W1 =WZDSGBSA=========(DBGHC====*=======)10MAR2011==16:39:02====MVDB2====D====2
GBP Db2 Read XI Read NF Read Changed Clean Sync Read Async Read
Name Target Hit % Miss Miss Writes Writes Total Total
BP0 DB1HC 14.0 3 40 12 0 50 0
BP0 DB2HC 22.2 1 13 20 0 18 0The XI Read Miss column now shows whether any members had to do additional I/O to read in pages lost through reclaims and cross-invalidation. Non zero counts are highlighted. From here, you can also hyperlink on a GBP to see interval and session counts for that GBP.
You have seen a few of the most important areas for data sharing monitoring. You have used a Single System Image context to enable you to monitor all members of a group, both individually and summarized, and to pull all of the relevant data together. Now you have the tools to master the new and complex data sharing performance issues.
- Return to EZDB2.