19.06.2013 Views

DB2 UDB for z/OS Version 8 Performance Topics - IBM Redbooks

DB2 UDB for z/OS Version 8 Performance Topics - IBM Redbooks

DB2 UDB for z/OS Version 8 Performance Topics - 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.

8.2.1 Per<strong>for</strong>mance<br />

We ran a number of benchmarks to try to understand the impact of the new locking protocol.<br />

Each test was per<strong>for</strong>med using an IRWW benchmark executing against a two member data<br />

sharing group.<br />

The tests were per<strong>for</strong>med with <strong>DB2</strong> V7 defaults and <strong>DB2</strong> V8 defaults. V7 uses LOCKPART<br />

NO as the default, while V8 behaves as if the table spaces are defined with LOCKPART YES.<br />

So, <strong>for</strong> some workloads <strong>DB2</strong> V8 would need to propagate more locks to the coupling facility<br />

than the same workload running in <strong>DB2</strong> V7. This is true if the workload running does not<br />

require all of the partitions to be GBP dependent. For these tests, all partitions were GBP<br />

dependent. So, the impact of LOCKPART(NO) in V7 compared with V8 should be minimal.<br />

We first review the impact on lock suspensions. Table 8-6 shows extracts from the Data<br />

Sharing Locking section of <strong>DB2</strong> PE statistics reports. The two columns under <strong>DB2</strong> V7 and V8<br />

represent each <strong>DB2</strong> member.<br />

Table 8-6 Protocol Level 2 - Statistics Report extract<br />

Suspends per commit <strong>DB2</strong> V7 <strong>DB2</strong> V8<br />

IRLM Global Contention 0.01 0.01 0.01 0.01<br />

XES Contention 0.54 0.52 0.00 0.00<br />

False Contention 0.13 0.11 0.00 0.00<br />

You can see from Table 8-6 that the benchmark produced significantly less XES contention as<br />

well as less false contention. This is because IX-L locks are now mapped to S XES locks<br />

which are compatible locks. The lock requests do not need to be suspended while XES<br />

communicates with the IRLMs to determine the IX-L locks are in fact compatible and can be<br />

granted. False contention is reduced because fewer lock table entries are occupied by X-XES<br />

locks. This is explained in more detail later in this section.<br />

Table 8-7 shows extracts from <strong>DB2</strong> PE accounting reports <strong>for</strong> the same test.<br />

Table 8-7 Protocol Level 2 - Accounting Report extract<br />

Class 3 suspend time<br />

(msec / commit)<br />

Lock / Latch<br />

(<strong>DB2</strong> + IRLM)<br />

This table reveals that, on average, each transaction enjoyed significantly fewer suspensions<br />

<strong>for</strong> lock and global contention. This is due to less XES contention, as a result of the new<br />

locking protocol.<br />

Now, we have a look at the impact on the coupling facility and XCF links. Table 8-8 shows<br />

extracts from RMF XCF reports.<br />

Table 8-8 Protocol Level 2 - RMF XCF Report extract<br />

328 <strong>DB2</strong> <strong>UDB</strong> <strong>for</strong> z/<strong>OS</strong> <strong>Version</strong> 8 Per<strong>for</strong>mance <strong>Topics</strong><br />

<strong>DB2</strong> V7 <strong>DB2</strong> V8<br />

6.219 5.708 0.080 0.109<br />

Global Contention 8.442 7.878 0.180 0.197<br />

Req In<br />

(req / sec)<br />

<strong>DB2</strong> V7 <strong>DB2</strong> V8<br />

2,300 2,200 17 18

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

Saved successfully!

Ooh no, something went wrong!