06.01.2013 Views

Download PDF - IBM Redbooks

Download PDF - IBM Redbooks

Download PDF - IBM Redbooks

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

The usage of some z196 features requires a minimum level of z/OS. For information about<br />

using the zEnterprise features and required software levels, see <strong>IBM</strong> zEnterprise System<br />

Technical Introduction, SG24-7832.<br />

3.1.3 Virtual storage constraint relief for DB2 10: Larger number of threads<br />

A critical issue especially for large SAP installations in the past was the constraint on DBM1<br />

storage below the bar (2 GB). Because each DB2 thread had a certain memory footprint in<br />

the DBM1 access space, this limit put a severe constraint on the number of acceptable SAP<br />

work processes per DB2 subsystem. This memory footprint is not static, but varies depending<br />

on the work done by the individual thread. In addition, in recent times, SAP applications tend<br />

to do more work asynchronously to the main thread so that they do not interfere with the<br />

application logical unit of work (LUW). As a consequence, SAP work processes open<br />

potentially several database connections, not just one of them.<br />

In effect, this issue required customers to closely monitor the actual DBM1 storage<br />

consumption and carefully assess the effects of changes in the application workload. Often,<br />

customers needed to go for n-way data sharing to map their SAP workload to enough<br />

database threads.<br />

DB2 10 places many data structures (such as the complete EDM POOL and thread working<br />

memory) above the bar, allowing for a massive scale-out of a single DB2 10 subsystem with<br />

SAP solutions. The result is less need to closely monitor DBM1 storage consumption. The<br />

result is also a large potential for DB2 member or logical partition (LPAR) consolidation,<br />

leading to less resource consumption and administration overhead.<br />

This feature alone is a significant contribution to total cost of ownership (TCO) reduction,<br />

especially for large SAP installations. It is already available in DB2 10 conversion mode.<br />

When planning to optimize the data sharing topology with DB2 10, in addition to the general<br />

TCO savings resulting from administering fewer DB2 members, fewer members can lead to<br />

needing fewer z/OS LPARs. In addition, if the members run similar workloads, less real storage<br />

for DB2 buffer pools is needed. Alternatively, with fewer DB2 members, the downtime of a single<br />

member can have a larger impact. First, the remaining members must handle the additional<br />

workload. Then, in a configuration with two DB2 members, if a member is stopped for<br />

maintenance, the remaining member is a single point of failure (SPOF). Overall, ensure that the<br />

performance of the remaining member does not degrade from effects, such as DB2 log and<br />

latch contention issues. This section continues by explaining efforts to tackle these issues.<br />

For a comparison of the virtual storage consumption of DB2 9 and DB2 10 for the same<br />

workload, see the white paper DB2 10 for z/OS with SAP on <strong>IBM</strong> System z Performance<br />

Report at:<br />

http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101845<br />

Savings of up to 93% in storage consumption can be achieved. Also, over 2,500 threads can<br />

be open at the same time.<br />

MAXKEEPD parameter: Due to this memory constraint relief, the SAP recommendation for<br />

the MAXKEEPD parameter has changed from = 12000. The MAXKEEPD<br />

parameter controls the number of statements in the thread local statement cache. Because<br />

the number of statements has been moved above the bar, no heavy restriction is on the<br />

storage that is used. Customers can now optimize the local cache hit ratio without<br />

compromising the maximum number of DB2 threads. Increasing the local statement cache hit<br />

ratio avoids implicit prepare statements, increasing throughput and minimizing response time.<br />

Chapter 3. Features of DB2 10 and the zEnterprise System as applied to SAP 31

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!